Marchete: or just recode all to use int pointers to big arrays
BrunoFelthes: yeah.. in one day i will not be able to do it... So, low legend is enough... no MCTS for me :/
BrunoFelthes: i think thta MCTS is the best search... what the top players are using?
Marchete: how do you manage simultaneous?
UndercoverToad: they don't tell
struct: :(
UndercoverToad: val growth tweeted something about DUCT
Marchete: in multi I'd try smitsi
struct: I read duct but I discarded it
Marchete: right now I can't make a GROW move score better than wait....
Kellthazar: I think that they are using some kind of witchcraft.
BrunoFelthes: I dont manage Marchete
struct: I dont have enough time to implement it
Marchete: how not Bruno
Marchete: MCTS must be some kind of simultaneous
BrunoFelthes: I do my play, and then the opponent play... If i use a seed space, the opponent cant use...
Marchete: either linear with sim each two levels
BrunoFelthes: it is close enough
Marchete: or DUCT/smitsi/UCT forest
Zenoscave: http://chat.codingame.com/pastebin/79558fe5-12e0-46e5-9134-40c208026c3b
PatrickMcGinnisII: BrunoFelthes i tried seed blocking, wasted my time
BrunoFelthes: but i think that simultaneous with intesection is too rare, it is not so hard to implement it at a normal MCTS tree...
Marchete: in C# Zeno?
Zenoscave: yes MArchete
Zenoscave: there's a syntax issue. I'm aware
Marchete: struct Move
{
unsafe public fixed ulong trees[4,2];
}
Marchete: probably, I should check at tech.io
UndercoverToad: shouldn't [,] be faster than [][]?
Zenoscave: No idea
Marchete: this struct is C++ like
UndercoverToad: [][] is an array of arrays
BrunoFelthes: the think that you can do simultaneous is seed and complete, that i think that is easy to manage... If you opponent chop a tree, just check the node before it, if it is a tree chop too, add one point...
Marchete: you don't need new
UndercoverToad: [,] is a 2D array, just an abstraction of 1 big array
UndercoverToad: less indirection
Zenoscave: gotcha
struct: For completes I store the gain on the action itself
UndercoverToad: [][] is called "jagged array"
Kiat: c# here i also using [,]
Zenoscave: And how do you populate one 2D array from another?
Whiskee: I have a precalculated bool[,,,,] Shadow for all sizes, please send help
Zenoscave: other = arr.Clone as ulong[,]; ???
UndercoverToad: Array.Copy I think
Kiat: ya
UndercoverToad: not sure if it handles 2D
Zenoscave: so not clone
UndercoverToad: no
Marchete: copy?
Kiat: copy faster
Marchete: Array.Copy probably, yes
Marchete: if you want super ugly C#
Marchete: http://chat.codingame.com/pastebin/abf34db9-0116-4512-82a1-557178fe92d9
PatrickMcGinnisII: BrunoFelthes i guess testing seed bounce is just a memcompare on boardstates
Marchete: m and m2 are different
Zenoscave: unsafe ugly is fine. Need a bit more sims I think.
Marchete: so m2=m; is deep copy
Zenoscave: got it. thanks
Zenoscave: for 2D array in Array.Copy I would do 8 elements not 4?
UndercoverToad: yes
UndercoverToad: just try it and see ;)
Marchete: also a lot of:
Marchete: http://chat.codingame.com/pastebin/4c27ca56-50cd-44a3-a031-72880a8611c4
Marchete: knowing how to copy, you can copy full structs as a whole
Zenoscave: why is there a void pointer in c# That's awful
Zenoscave: I don't like that.
Zenoscave: I feel gross
Marchete: :D
Marchete: it's dirty
Zenoscave: very
Marchete: but instead of manually copying 20 fields
Marchete: ok ok
Marchete: nvm it's really ugly
Zenoscave: If only I could figure out how to make a struct work with my Fast Priority queue. I need a class sadly
Zenoscave: but this giives me enough to improve I think
Marchete: class { struct k;}
Marchete: but it's ugly too
Zenoscave: you are getting too good at ugly C#
Marchete: I'm corrupted by C++
Marchete: before C++ I never liked pointers or arrays or similar
Marchete: so don't do it
Lysk: CouscousAgha I just pushed you
Lysk: also 5H4dy
CouscousAgha: I love you Lysk
Lysk: :blush:
CouscousAgha: i was tempted to resubmit
CouscousAgha: and you came with the clutch
Lysk: it was my pleasure, I hope I get to legend too
CouscousAgha: im cheering for u
ClockSort: aaugh i resubmitted and missed the help of Lysk
CouscousAgha: Lysk clutched it
Lysk: tiraptor also helping!
PatrickMcGinnisII: http://chat.codingame.com/pastebin/d05f96f9-bfbf-4a1e-b2c4-c337081dad5f
PatrickMcGinnisII: does that work?
ClockSort: :eyes:
PatrickMcGinnisII: i guess not
PatrickMcGinnisII: been awhile
Lysk: yay
CouscousAgha: welcome to legend Lysk
Marchete: see? that's pure evil uglyness
PatrickMcGinnisII: :thermometer_face:
CouscousAgha: This is my first ever CG challenge and I managed to get to Legend with only heuristic
CouscousAgha: I think that's a cool achievement
Kellthazar: Indeed
UndercoverToad: and now even another toad entered the game
Marchete: who?
UndercoverToad: around rank 20, daiver
UndercoverToad: known from topcoder
Marchete: ahh
Marchete: it's like the evil toad
UndercoverToad: according to ceg i'm the evil toad
UndercoverToad: so that guy must be the good twin
Marchete: like in evil bart :D
UndercoverToad: but then i'd have a giant scar on my side
Marchete: :eyes:
aCat: morning euler ;-)
UndercoverToad: hi
aCat: yesterday I couldn't do anything with code
aCat: today I should have time
aCat: final straight
UndercoverToad: done with reviews?
aCat: yeah
aCat: 6 in one day
aCat: damn
UndercoverToad: that was fast
UndercoverToad: my prof needed at least to months for my master thesis
aCat: I've read some papers before but didn't wrote the revir=ews down
aCat: papers are short ;]
aCat: and most of them crap
aCat: one was good
UndercoverToad: so just one to be published?
aCat: and 2 out of 4 reviewers were negative talking about author doing something with MCTS
aCat: while the paper was NOT about MCTS but NRPA
UndercoverToad: what's wrong with MCTS?
UndercoverToad: ah. would have to read about NRPA to see how far that is off
aCat: th point is there were 'confident' reviews that didn't get that the algo is different
aCat: quite
aCat: single player
aCat: it uses monte carlo pleyauts ofc
aCat: but otherwise meh
aCat: no idea about the final decisions made by the chairs yet
aCat: regard=ing acc/rejection
aCat: but really the reviewing system is strange / unfair and really random
UndercoverToad: because reviewers aren't always experts in that field?
aCat: thets only a pertial problem
aCat: you evaluate your confidence, not only score
aCat: http://chat.codingame.com/pastebin/ea567375-08c5-48f6-a382-4aa23079bcaa
aCat: but people tend to claim they are expersts
aCat: and also have different meaning of scores
aCat: you san see very similar review as text
aCat: and oune sais weak accept, and other strong reject
Zenoscave: I have a very flat scoring system for inner day moves I realized
Zenoscave: Most of my top scores are equal as long as the same moves are applied in a day
Zenoscave: What's the best way to get around this? turn decay?
Numby: top player in legend is 10 score above 2nd O.o (reCurse at 42, 2nd player at 33)
Numby: O.o
Zaphus: oh, look at that... when I stop accessing arrays past their bounds, I jump from ~800th to ~500th :-)
I can't even work out how I got that high with such broken code!
miklla: reCurse vs the world : 142 - 5
Numby: that's insane I thought the game was super close at top positions and everyone plays close to optimal
Zenoscave: Close. but recurs_e is closer
JohnCM: i've come to ignore reviewers, especially those who give irrelevant comments
JohnCM: does more harm than good listening to non-constructive feedback anyway
JohnCM: hmm... how to beat gold boss with heuristics
JohnCM: i'm at 10 pts difference from gold boss alr
JohnCM: now just need to code some useful stuff
Westicles: Wow, team North America takes a big lead
NinjaDoggy: wait is this real?
NinjaDoggy: reCurse is 9 points ahead of second? -_-
NinjaDoggy: reCurse explaing yourself -_-
NinjaDoggy: explain*
aCat: JohnCM you can ignore them, but if they decide your paper is rejected and you have nowhere to shout that they make mistakes it is frustrating
JohnCM: indeed, but you got to believe in yourself. surely there's another venue to submit to
Zenoscave: I really broke my bot. I can't find the right eval anymore
NinjaDoggy: Recurse is like an alien racing coming to earth and teaching us plebians how to really play chess :)
JohnCM: go history and get your previous code
aCat: this si problem if youre forced to aim for top conferences
aCat: otherwise its not much a big deal
Zenoscave: JohnCM There's 201 submits...
Zenoscave: I can't remember which was best
JohnCM: wow. time to start from scratch, maybe you'll code something better
Zenoscave: Seeing how I'm in legend I doubt it ;)
NinjaDoggy: I'm going to recode after watching a few hundred recurse replays ;)
JohnCM: haha legend is good enough tbh
Zenoscave: not enough time left in competition
Marchete: WARNING: Illegal reflective access by com.google.inject.internal.cglib.core.$ReflectUtils$2
JohnCM: i have a feeling recurse didn't do heuristics
NinjaDoggy: the game is supposed to be played completely differently apparently ;)
Marchete: what's that error in my mvn brutaltester?
NinjaDoggy: I feel like the only way was he did some NN stuff :/
Zenoscave: what jdk are you using?
NinjaDoggy: otherwise gap shouldn't be this massive
JohnCM: if NN will be quite impressive
JohnCM: you have to test it locally and then port the weights over
Westicles: Naw, probably smitsimax with lots of booking
NinjaDoggy: no way it's smitsi...
NinjaDoggy: maybe simul MCTS
JohnCM: haha if codingame has a tensorflow/pytorch support library, i would use NN too
Marchete: it seems the referee
JohnCM: but don't think codingame wants to go that direction
Numby: prolly too many people would go for NN if there was an easy way to do it
Numby: yeah that would be not as interesting
Marchete: hmm, why not?
Zenoscave: They had mnist as a medium problem once upon a time
Marchete: I mean
JohnCM: i actually kinda like the fact that heuristics can get gold
Marchete: the point is creating the best AI bot
NinjaDoggy: who has the best muZero clone contest ;)
Zenoscave: Yup. My bot'll be bottom barrel legend :/
Zenoscave: probably shouldn't be in legend at all
NinjaDoggy: legend is legend :)
NinjaDoggy: no difference until top 20 :)
Zenoscave: fair
NinjaDoggy: and it's never over until the last day :D
SterlingA: Do you guy know any heuristic bot in legend ?
Zenoscave: I'm exhausted. I've done easily 100 hours
JohnCM: this comp i took a diff strategy. i coded till bronze, took a weeks' break, then came back over the weekend
JohnCM: good to have the silver and gold bosses ready for the benchmarking
Zenoscave: I keep forgetting to eat
PaarthThePro: No way!
Marchete: Zenoscave, do you know what can cause this?
Marchete: http://chat.codingame.com/pastebin/aadf2db5-c87b-4f74-a478-23e7528e77b7
UndercoverToad: NinjaDoggy 08:10AM
Recurse is like an alien racing coming to earth
Val Growth agrees
Marchete: on referee or mvn or JDK11 or whatever
UndercoverToad: https://twitter.com/valgrowth?lang=en
ddreams: Marchete too new java
ddreams: get java8
NinjaDoggy: you link me his twitter but I can't find a tweet about recurse :(
Marchete: thanks ddreams
NinjaDoggy: it's also not translated to english :(
ddreams: wow 41.13 vs second place of 32.79
ddreams: that's impressive
Marchete: I love how Java is so incompatible between versions. C# lacks that and I'm furious
UndercoverToad: re curse was called an "extraterrestrial" if I can trust deepl
Westicles: Ask him his approach and he'll be as forthcoming as one of those arrival hexapods
Westicles: *heptapods
Marchete: on brutaltester "Negative score during game 1 p0:-1" ?
UndercoverToad: a crash
UndercoverToad: or timeout
ddreams: start the game without brutaltester and see
ddreams: depends on the referee, you may have one that does not give you possibleactions
ddreams: or you're playing in wood league
JohnCM: if only codingame has a library like OpenAI, where you can env.step() all your actions without coding out the entire environment haha
JohnCM: then monte carlo will be easier
UndercoverToad: how does that work with local testing?
Marchete: it seems crashes, yes
Marchete: but if I run on VSstudio
Marchete: there is none
UndercoverToad: i tend to have an
- if DEBUG
break main loop after first turn
Marchete: nvm
Marchete: I'm more stupid that I can even imagine
Marchete: I compiled DEBUG and not release
OBenjOne: I had almost no time to work on this challenge... It would be nice to make it to legend though
UndercoverToad: hah, I got what's wrong :)
Marchete: and release was indeed crashing
Marchete: yeah
UndercoverToad: well, almost
NinjaDoggy: on a side note, why use brutal tester instead of own arena? :P
NinjaDoggy: (assuming you already have full sim)
OBenjOne: what's brutaltester? Sorry I'm new to this
OBenjOne: Should I be using it?
UndercoverToad: program to help you test offline
UndercoverToad: play a bunch of games, 2 program versions vs each other
OBenjOne: That seems really useful, I'll look into it
PatrickMcGinnisII: i can't believe i'm still finding bugs
UndercoverToad: the more bugs you find, the more likely it is that you still have some more remaining
PatrickMcGinnisII: was gonna expand depth, but bug would just make it worse
dbdr: so there are infinitely many bugs UndercoverToad? :D
PatrickMcGinnisII: toad²
UndercoverToad: don't use logic against me :D
UndercoverToad: just quoting pretty much the only thing I remember from my software reliability class
dbdr: :D
dbdr: reminds me of: for cheeses with holes in them, the more cheese you have, the more holes you have. and the more holes you have, the less cheese you have. so the more cheese you have, the less cheese you have
PatrickMcGinnisII: so cheezy
ddreams: makes sense
ddreams: kind of like the companies that tell you "the more you spend, the more you save!"
PaarthThePro: I like that quote
PaarthThePro: The more I live the less I live
PaarthThePro: I am immortal
PatrickMcGinnisII: ouch
PatrickMcGinnisII: ok, rock out with your chicken out
PatrickMcGinnisII: :poultry_leg:
PatrickMcGinnisII: i think that hurt my ranking
shitizen: 冲国人？
PatrickMcGinnisII: :baby_bottle:
Westicles: something middle something man?
ddreams: I think he's looking for other chinese people
ddreams: guessing based on google's guess
PatrickMcGinnisII: well my rank is better by fixing the bug, that's a first
ddreams: board with ten holes: seed=8472451449982904300
ddreams: you may be surprised to learn that rebless crushed me on it
ddreams: actually 132 to 129, but I suspect it wasn't really close
shitizen: 你们上班要996吗
shitizen: you are so free
ddreams: you are not?
shitizen: do you start working at 9am to 9pm for six days?
ddreams: often, but not because I am forced to
ddreams: I get paid per hour, so I like to work a lot in periods and then take time off
shitizen: that's a good idea
UndercoverToad: I work 7:15am-3pm
Bari: Hi
NinjaDoggy: reCurse's sun effeciency early game is actually insane :O
PatrickMcGinnisII: can't see, post replay
shitizen: how do you learn english
my english is so poor and bad
ddreams: when I was a kid, I played English games looking up everything in a dictionary
shitizen: one word often include many meaning
ddreams: practice
shitizen: it's too hard to remember
shitizen: i often look up the words that i don't know but after a while i forget what it is
ddreams: language has to be used, or you'll forget
shitizen: you sent "take time off" just now.but i didn't know what it is until i translate it by online translater
ddreams: well, now you know :)
Michael_Howard: NinjaDoggy what does he do? Please can you post a reply for us poor Goldies who can't see up there?
UndercoverToad: https://www.codingame.com/contests/spring-challenge-2021/leaderboard
UndercoverToad: of course you can see it
UndercoverToad: "view last battles"
Michael_Howard: Thanks, I forgot you could do that on other leagues :)
shitizen: but english still have great many phrases that i even don't see them
PatrickMcGinnisII: ok finally in top 1k, i can go to bed
PatrickMcGinnisII: ahh yes, tx UCT
LelouchVC2: I'll make 1000 accounts and make them all top 1000
LelouchVC2: Just to deprive you of that pleasure
ddreams: too late, he's already dreaming about it
LLG: ClockSort boutta hit legend
jrke: wut reCurse 1st only under 24 h
jrke: with 8 points lead
jacek: legend on 1 submit or delete
ddreams: #1 on 1 submit or delete
jrke: is that NN
UndercoverToad: we don't know
jrke: but he made rank 1 under 15-16 h think
UndercoverToad: but it's to be assumed that he was testing with an alt account before. he knew that this would be a #1 submit
jrke: even less than that
UndercoverToad: an impressive performance nontheless
UndercoverToad: he won 2 contests before. but this is by far the most dominant win I can remember (among all contests, not only among his wins)
jacek: if server kept down that would be quite a fail for him
Westicles: not a win yet
kovi: no way to beat nn
jacek: i bet it is NNish
UndercoverToad: the only thing that can stop him now is staff
jacek: or some 9mms
jrke: http://cgstats.magusgeek.com/app/spring-challenge-2021/reCurse
UndercoverToad: nice, I even got the honor to play against him
UndercoverToad: dropping his score by losing to him probably
Westicles: Staff? Like obfuscation ruling?
UndercoverToad: i don't think it will happen
UndercoverToad: just a deus ex machina that's more likely than any player beating him in the arena
jacek: a modified game to make it sequential with known moves
kovi: i was afraid of this gonna happen. game is too "simple", single search + "heuristics" (which within search can be best captured by nn)
it kinda takes away motivation...today...and for long term as well
jacek: thats why he waited so long :?
davilla: are alt accounts legit here?
derjack: ban
dbdr: no
UndercoverToad: they are not? :scream:
dbdr: :D
jacek: at least dont submit in contest
dbdr: yeah
VirtusDomo: god its hajrd to get momentum
VirtusDomo: i swear i'm always stuck
ddreams: ask recursive for tips, virtus
ddreams: he's good at getting momentum
VirtusDomo: recursive>?
davilla: well there's your deus ex
davilla: banned account
VirtusDomo: lol is that like a known thing here?
VirtusDomo: im new
ddreams: top player miles ahead of the rest
VirtusDomo: ahhh i see i see
ddreams: in one submit
VirtusDomo: yeah i'm just trying to get good enough for internships lol
davilla: I wouldn't ban the guy honestly, but maybe force him to use the other account
davilla: well now I don't feel so bad for dropping out during wood
ddreams: true, wood leagues were challenging
davilla: it's more that I couldn't get my head in it, motivationally
davilla: my program actually made it to silver without any further mods
davilla: don't ask me how
ddreams: I'm in legend and just found out I calculated resource points incorrectly
davilla: sorry, I meant bronze
ddreams: maybe fixing it will put me in top 100
davilla: cool :)
ddreams: anyway, gtg family stuff, bbl
Selwyn: man I feel like I'm an intermediate developer but I can't even do any challenges here
Selwyn: I feel like I'm so badd bahaha
VirtusDomo: lol same man same
VirtusDomo: even the easy ones have me googling
Selwyn: sameee
thanos98: Hi guys anyone has some materials about the challenge code4life?
ikustom: tes.
ikustom: yes*
thanos98: Can you share it?
thanos98: ???
UndercoverToad: https://www.codingame.com/forum/t/code4life-feedback-strategy/2835
happy reading thanos98
thanos98: Thx!!
Scarfield: i see both smitsi and zeno made it in legend :muscle:
Scarfield: Rebless using NN or just chat guessing?
jacek: how else he would lead that much?
Scarfield: secret sauce, idk
Scarfield: jacekMax maybe
jacek: that would give him 100% win
jacek: it cant be that
Scarfield: xD
DomiKo: Wait
DomiKo: what
UndercoverToad: you opened the leaderboard?
DomiKo: yes
UndercoverToad: you spilled your morning coffee?
DomiKo: if I had one
UndercoverToad: good, coffee tastes horrible and has a disgusting smell
Scarfield: :scream:
DomiKo: some smell goood
DomiKo: OK I run 3 games with him, and he beats me +2-5 points every game
DomiKo: like how
UndercoverToad: like he's not even trying
DomiKo: maybe it's his worst version
jacek: https://www.reddit.com/r/cats/comments/hfjr6r/he_buries_coffee_because_he_doesnt_like_the_smell/
jacek: win is a win
dbf: with 5 ms timelimit
Noyotens: https://www.codingame.com/replay/556574644
Noyotens: what i need to make my code 10/10
VizGhar: https://www.codingame.com/replay/556546205 hmm timeout or wroung output?
jacek: suicide
jacek: maybe he cant handle simultant seeds
Astrobytes: reBless, wtaf
MyCCODE: I wanna compete
MyCCODE: I wanna compete
MyCCODE: I wanna compete
MyCCODE: I wanna competeI wanna competeI wanna compete
UndercoverToad: "I wanna get kicked"?
MyCCODE: http://chat.codingame.com/pastebin/c66e7c93-5abc-47fc-bb8e-c1aaae67dbe0
UndercoverToad: Astrobytes would you be so kind? ;)
MSmits: it makes sense Astrobytes, a simple positional game like this seems right for an eval with NN. Just the input seems a bit hard to structure
Astrobytes: My pleasure Toad.
Astrobytes: MSmits. Yeah, I was thinking this. Will be interesting to hear about it.
MSmits: we all find that this game is hard to eval, nn takes care of i
MSmits: it
MSmits: not saying that it's easy write a nn here, just saying it makes sense to do it if you can
UndercoverToad: or it fails hard https://www.codingame.com/share-replay/556411886
Astrobytes: Yes, I agree MSmits
UndercoverToad: but in 99% it's just overwhelming
MSmits: weird that it would fail so hard
dbdr: maybe it predicts a loss
MSmits: yeah, all output close to 0, one loss being much like another, just by a different margin
Astrobytes: Either way, I'm quite impressed
MSmits: same
MSmits: but not completely surprised tbh
dbdr: that's why I find it more satisfying to optimize the margin
Astrobytes: Yeah, same here Smits
AntiSquid: are you sure it's NN ?
dbdr: but maybe win/loss has advantages
MSmits: AntiSquid pretty sure
MSmits: reCurs e doesnt do anything else these days
MSmits: and the margin is insane
Astrobytes: Which we know about from the other board game leads
Astrobytes: *that he has/other NNs have
AntiSquid: cool, then which type of algo, was looking at ReBeL just because of this contest
MSmits: I do feel a bit sorry for the people leading the leaderboard before recurs e took it
MSmits: i know the feeling from being wrecked by nn's on board games
MSmits: bit hopeless
Astrobytes: All's fair in love and CG Contests
MSmits: sure, i just feel sorry for them i sall :)
MSmits: empathy and all that sht
kovi: thx
Astrobytes: Oh for sure!
AntiSquid: but . there's a HUGE chance it's not NN, for the simple fact that all you need to do is find right heuristics, so theoretically with decent data analysis you can just figure the right heuristics out
AntiSquid: to get to legend at least
UndercoverToad: i still don't get the hiding and last minute submit. isn't it a more fun experience without
UndercoverToad: ?
MSmits: UndercoverToad probably he wanst sure his nn would succeed
MSmits: a poor nn migh tnot even get out of wood
AntiSquid: hide until 1st place is secure lol
MSmits: they are kind of all or nothing in a way
UndercoverToad: i'm pretty sure he knew before the servers died yesterday
MSmits: pretty sure he would have submitted if it would end top 10 :P
MSmits: instead of 1
UndercoverToad: i'll just wait for the post mortem instead of guessing any further
MSmits: yeah, sure
Astrobytes: Yeah. Should be a good read.
LLG: e
MSmits: gotta code some more, somewhat satisfied with top 100, but can do better maybe
Pick8chu: Hey MSmits
MSmits: hey Pick8chu, was just heading out
AntiSquid: my win rate just drastically improved by adding 1 extra condition ... if i guess few more i will make it to legend with just that .
Pick8chu: oh
dbdr: "maybe"
Pick8chu: alright good luck man and congrats on 100
MSmits: thanks, ttyl :)
Pick8chu: :)
Pick8chu: Would anyone give me some thoughts on improvement?
Pick8chu: https://www.codingame.com/replay/556597597
AntiSquid: add random heuristics and test
UndercoverToad: seed on 9 is in your own shadow range
Pick8chu: is it?
UndercoverToad: wait, i was looking at blue
Pick8chu: haha okay
UndercoverToad: seed on 6 then
UndercoverToad: 15 too
Pick8chu: How could i apply random heuristics?
UndercoverToad: when the tree grows to size 3, you have a problem
Pick8chu: I complete them too early?
Butanium: Hey
UndercoverToad: yes, was about to say that
Pick8chu: AntiSquid How could i apply random heuristics?
therealbeef: true on NN all-or-nothing. I'm doing NN for eval with search (for the first time), and was stuck in 1500th in silver until 1 good update gained 2000 places in one go
UndercoverToad: and the positioning at 1 and 4 isn't optimal again
Pick8chu: 1 and 4 you meant turn?
Butanium: I want to do MCTS for the challenge and I'm thinking about making playout returning none discrete value depending on the score delta between the two players
UndercoverToad: 2000 places doesn't mean much in a contest this size ;)
NotSureWhyThisWorks: UndercoverToad you used beam search right?
Butanium: do you think that's a good idea?
UndercoverToad: cell 1 and 4
UndercoverToad: turn on debug mode
MUkhtadirshaikh: hello word
MUkhtadirshaikh: world
UndercoverToad: yes, beam all the way
Pick8chu: they have debug mode???
UndercoverToad: settings, gear icon
MUkhtadirshaikh: yup they have it
Pick8chu: lmao
UndercoverToad: even mentioned in the statement ;)
MUkhtadirshaikh: lets clash
NotSureWhyThisWorks: did you have to limit the possible moves so that it was possible?
UndercoverToad: i'm in this debug mode the whole contest already
dbdr: there is a non-debug mode?
UndercoverToad: some limitations on spawns to generate less states
UndercoverToad: not going too much into detail during the contest
NotSureWhyThisWorks: yeah fair enough
NotSureWhyThisWorks: will there be post-mortems after the contest finishes? I am interested to see how the legends did it
UndercoverToad: yes, there are always some
UndercoverToad: not sure if i'll write anything yet. my bot doesn't have any interesting thoughts in it
jacek: so you copied your brain into it?
NotSureWhyThisWorks: surely a nice eval function or something
NotSureWhyThisWorks: you are 11th
Astrobytes: offf buuuuurn
Astrobytes: jacek bringing the cheek
UndercoverToad: that was inappropriate jacek
jacek: :(
Astrobytes: /kick
UndercoverToad: i extended my board state and realized I forgot to add this field to hash+equals
UndercoverToad: when adding it, the offline batch gets worse
Astrobytes: lol
UndercoverToad: maybe because the beam gets more narrow that way
Pick8chu: https://www.codingame.com/share-replay/556609594
Pick8chu: Slightly better
NotSureWhyThisWorks: I realised when I was pruning I wasn't sorting beforehand so I changed it and it got worse
jacek: username checks out
BrunoFelthes: OMG reCurse started the contest with 94% win rate
NotSureWhyThisWorks: lol
BrunoFelthes: this guy is insane
UndercoverToad: no, he didn't. his wood bot wasn't impressive at all
Butanium: do you think I could do any improvement on my heuristic?
Butanium: https://www.codingame.com/share-replay/556555961
Butanium: I'm about to switch to a MCTS but I won't have the time to siubmit it before the end
Butanium: I'd like to end up in legend x)
AntiSquid: then do heuristics instead LOL
Pick8chu: AntiSquid how do you do that tho?
Pick8chu: like generate random numbers and see if what gives you the best result?
UndercoverToad: random actions you mean?
Pick8chu: heuristics
UndercoverToad: that can be an option. you still have to define how a good result looks like
UndercoverToad: then i don't understand
UndercoverToad: back to my code
Pick8chu: how do you mean? you define what a good result is?
Butanium: AntiSquid: Do you think I could improve my evaluation function without doing random search ? x)
AlMualem: guys what is the strategy to win the silver bot...(complete as fast as you can,complete the ones with the sallest index for more points,seed with distance >=2 each tree from another...)any other thoughts?
AlMualem: smallest index*
particleflux: lol only just now realized that richer soil == smaller index
cegprakash: reCurse 93% win rate in legend wow.. Nice way to troll everyone on legend. Looks like he cracked this contest.
AlMualem: hahaha
Pick8chu: that's jsut crazy lol
AlMualem: i have implemented all these thoughts to my code and still can win silver bot in every match :(
AlMualem: cant*
actualCoderTrevor: Do you consider shadows?
lionandmouse: try to maintain some trees for stable income in good positions and then farm. Shadows can be disregarded until gold.
actualCoderTrevor: You can only get so far without doing that
Sanguigno: lca
AlMualem: i took them (the shadows) a little bit under consideration,,,i mean that is the reason i am seeding with distance greater or equal than 2
TootiFlooti: what division are yall in
AlMualem: whats left is to implement , is seeding in the smallest possible index
TootiFlooti: im still working on getting to silver :(
AlMualem: i am 222th on silver mate
TootiFlooti: nice
TootiFlooti: i have a stable build and just waiting tto get thru bronze im beatingf everyone so far but sadly i was not able to work on this all week :(
AlMualem: i am working as we speak :P
TootiFlooti: im going to try to beat you almualem thats my new goal lol
Pick8chu: AlMualem as far as I know, everyone here implemented with weights. Basically, among possible actions, what is the best option
TootiFlooti: im to sleepy now ill work on it tomorrw
TootiFlooti: goodnight all
AlMualem: bb
Pick8chu: good night man
AlMualem: pIckachu? when we talk about best actions...regarding what?
Pick8chu: let's say on one day you get grow and seed and complete and wait
TootiFlooti: aslo quick question is there a prize for winning contest
TootiFlooti: im jsut doing it for practice
Pick8chu: among 4 of them, you weight them using the information you have like sun, or shadows, or how much sun point i have
jacek: a dinner with chosen mod
Pick8chu: and pick the biggest weight
TootiFlooti: ty jacek
k4ng0u: if you manage to beat reCurse, you will likely earn the respect of the whole community
TootiFlooti: lmao
Pick8chu: lol
Astrobytes: jacek :rofl:
TootiFlooti: true i like that. respect like fast and furious classic
AlMualem: pica pica i changed a number and now i have won the last 4 game scenarios against silver bot...that number indicated the number of trees i shoulf have on deck...and of after a certain day to stop stop seeding and try to grow and complete whats left
michaelzlatkin1: How do you even get out of the first bronze league? It really just feels like the winner is determined by who gets the luckiest bump...
KelvinAndHubbles: Woot Woot! 4 days hardstuck top 100 Silver, makes some tweaks every day and friken finally past it
peerdb: michaelzlatkin1try seeding your new plants in gridlike patterns, so they dont block each others sunlight
Pick8chu: Good job man!
Pick8chu: KelvinAndHubbles congrats
michaelzlatkin1: plants...?
AntiSquid: https://www.codingame.com/replay/556639948 2nd win, Automaton2000 give me some dumb ideas
Automaton2000: how can i send a private message
peerdb: click on someones name in chat here Automaton2000
Automaton2000: i'm not a fast coder
AlMualem: https://www.codingame.com/replay/556638953 pic8chu this is my last battle with boss
Whatar: it doesn't matter, we all know that you are the real OG Automaton2000
Automaton2000: i'm pretty sure it is the same as the other ones
AlMualem: i am testing it right now in arena
peerdb: Is there an easy way to check where the segfault happens in my code (c++)?
Pick8chu: AlMualem looking good
AntiSquid: https://www.codingame.com/replay/556639948 making my bot dumber really helps
CarbonBlack: Hi
Pick8chu: i don't get why edge has benefits
AlMualem: AntiSquid i see you are seeding a lot on the outskirts and win...first of all gj mate...second you thing its more important to complete first or complete (smaller index=more points?
Pick8chu: *edges
AlMualem: think*
Mistblinde: Less shadows on the edge
geppoz: edge of course have at least 1/2 of "a week" without shadows
geppoz: mid can have all week shadow
AlMualem: less shadows --> more sun points--->complete first...right?
Pick8chu: hmm okay let me try that
peerdb: yeah I was using my trees on the edges to farm sunpoints and the trees in the center to farm score
nam4.nguyen: My first algo reach top 1k but after modifing something I feel make more sense, my new bot got worse rank. Idk why my first stupid bot can be that good :|
Pick8chu: actually that's very good idea peerdb
peerdb: :nerd:
BlitzProg: I'm looking at mcts tutorials but there's something like "win amount divided simulation amount"
BlitzProg: can I just replace "win" by something like my score?
BlitzProg: maybe divide it by a number to keep it between 0 and 1
AlMualem: my strategy is to first reach 6 trees at level 3 and complete the 2 with the smaller index(with this strategy i farm the ones closer to the center and let the rest to farm...after a certain day of the game i complete every tree i can starting from the ones that are already 3 level
AntiSquid: i don't understand the questions, i just added some random heuristics
Pick8chu: How do you do hat
AlMualem: with my last submit i went from 222nd to 76 and still going (i am at 72%)
AlMualem: until now i haven't battle the boss :/
AlMualem: 56th at 76%
MSmits: BlitzProg you can do that
AlMualem: 22th at 79%
MSmits: you dont need to keep it between 0 and 1 either, no point to that
MSmits: just adjust your exploration parameter
MSmits: until it explores properly
BlitzProg: MSmits thanks
20 hours lefts, never tried mcts but i'm so desperate lol
MSmits: if your scores are between 50 and 100, your exploration parameter should be of the order of 50 or so
MSmits: same if your scores are between 450 and 500
MSmits: it's the score interval that matters
MSmits: the size of it i mean
BlitzProg: Oooh I see
MSmits: but you do need to fit it, i am just giving you an estimate
AntiSquid: Pick8chu, think of a random condition and write that into code. Here's the pseudocode:
"if (something is true) then do this action."
CrystallineCore: In coders strike back, sometimes my bot revolves around the same checkpoint multiple times before going to the next checkpoint. Is this the problem of my code?
Pick8chu: Like super random?
MSmits: CrystallineCore yes, you are overshooting it
Pick8chu: Like if(tree_size = 2) seed?
BlitzProg: Well, here goes nothing then, I'll spend some hours coding that and see what happens
MSmits: you need to reduce speed or aim away from the checkpoint
MSmits: to correct for overshooting
MSmits: BlitzProg mcts is pretty hard to debug, at leas that is my experience. It might take a while to start workinf
MSmits: unlike something like beamsearch which could have some reasonable results fairly quickly
Mazelcop: hello folks
BlitzProg: Beam search requires an eval tho
Mazelcop: anyone knows why my local CG referee starts each player twice ?
MSmits: BlitzProg yeah
MSmits: well mcts does too, if the depth is too great
AlMualem: DAAAAmn i finished 3rd....
MSmits: how much rating below boss?
AlMualem: and i played only once against the boss should i try again with the same code?
peerdb: just wait AlMualem , you might get bumped up a spot or 2
peerdb: what is the boss' arena score vs yours?
AlMualem: the boss 43,59 me 42,79
MSmits: if you are 2 pts or more below boss, you have no chance of that
MSmits: that might be doable
MSmits: resubmit i'd say
peerdb: id wait 15 minutes
MSmits: sure yeah
wlesavo: depends on league
AlMualem: why wait peerdb?
BlitzProg: Stakes are higher for me this contest so I'm pretty willing to try many things to get into legend, but it has been hard
peerdb: when other people play their matches they also play you, so it impacts your score.
AlMualem: 43,62 the boss 42,96 me
Nerchio: Mazelcop i didn't notice anything like this but i was timeouting a lot so maybe i have same problem i just didnt notice? :D
BlitzProg: Since Monday nothing did better than what I sent that day
AlMualem: ok i will wait
peerdb: same goes for the boss,
so you have a chance to win some matches while the boss loses and you go above him AlMualem
AlMualem: i will wait 10 minutes
peerdb: sure
AlMualem: and submit the same code again?
peerdb: thats up to you
Sanguigno: elca
VRtheKing: how to clear the quest called moderate a contribution
AlMualem: if i submit the same code is there a chance to promote? i am 4th at the moment....43,36 the boss 42,96 me
MSmits: there's a chance
MSmits: though hard to say how much of one
MSmits: it depends a lot on other players and luck
peerdb: perhaps you can get slightly better results by changing your strategy just a little (harvest earlier/later)
AlMualem: i will submit one more time with the same code and if i dont promote i will test when its better to seed
AnthonyHilala: comment voir le classement du boss ? :s
peerdb: okay good luck AlMualem
AlMualem: mt difference with boss is 0,4
AlMualem: ty mate
peerdb: 0.4 is not much
AlMualem: sorry for spamming i am ultra hyped right now
ItsNotABug: wow, recurse has smashed legend
peerdb: AlMualem how far below boss now?
cegprakash: ppl in top legend be like.. oh noo it's time to switch to NN
AlMualem: i promoteeeeeed automatically :D :D :D :D :D :D :D
peerdb: whoooo
peerdb: congrats
AlMualem: ty mate <3
AlMualem: i dont have a lot of experience in coding...but i am really into coding want to be a full stack developer one day
Neon_Tuts: Hello World
Zanoshky: Hello Neon
AlMualem: hiiiiii
The_ZimaBlue: can you guys check out:
https://www.codingame.com/contribute/view/67177d35db600c4defaf302f9fdeea1f6c3f
cegprakash: what is the maximum number of possible actions that we can get in input
cegprakash: numberOfPossibleActions has any upper limit
AlMualem: there must be a max number because our map i limited but how can this be helpful?
cegprakash: I generate possible moves and I care about memory :P
cegprakash: I used 300 and it crashed some games
solistice: 37 * 7 + I guess
cegprakash: so I increased to 2000 now
solistice: that's without thinking about stuff
solistice: think 300 should suffice
solistice: but that's not at all a calculated value
AlMualem: why youdo you generate and not reading the input from possible_action list?
cegprakash: for future turns
cegprakash: for opponents
cegprakash: things like that
AlMualem: hm...i see...ok...so lets calculate it...
AlMualem: you said you need it for the opponent move,not yours...right?
cegprakash: for ours too during future turns
cegprakash: when u simulate the future like Doctor Strange
AlMualem: ok ok for both
AlMualem: ahahaha
cegprakash: u can look 2million possible outcomes
cegprakash: and choose the path with 1 winning outcome
AlMualem: do you want to take under consideration the sun points? (i suppose yes)
UndercoverToad: of course
Pick8chu: So you try all the possible outcomes till your memory allows and decide on that?
TobiasA: the only way for your bot to win is for it to use the infinity stones to snap the opponent away and die in the process
Pick8chu: lol
TobiasA: it really works
Pick8chu: but how ? that's just crazy
TobiasA: doctor strange said it is the only way
TobiasA: strange never lies
Pick8chu: so you can have 7 boards so start try 7 possible ways?
Sanguigno: elca
solistice: 14 million possible outcomes
solistice: only 1 is winning
TobiasA: my bot found the reality stone:hushed:
Pick8chu: could you help me to understand?
UndercoverToad: there are many different ways what to do with the simulations. You have to ask the right question for a meaningful answer
AntiSquid: Pick8chu what's your job role at samsung ?
TobiasA: first harvest all your trees and craft a gauntlet
Pick8chu: web developing for now, as a full stack developer
Pick8chu: I get how you should weight the possible actions
AntiSquid: cool
Pick8chu: but I don't get how to simulate with those actions and see what brings the best outcome
TobiasA: wow you work at samsung
AlMualem: so...depending on the number of the trees you own and their size.,,, http://chat.codingame.com/pastebin/0f136a7d-8e71-4c78-b35d-590fd38684f8
UndercoverToad: usually you don't weight actions but states
UndercoverToad: start with a board, create all possible following boards
Pick8chu: oh
UndercoverToad: then rate these based on some criteria (game points, suns, ...)
AntiSquid: Pick8chu try this page first : https://www.chessprogramming.org/Best-First
UndercoverToad: remove a few if there are too many (would get too slow)
UndercoverToad: and further expand the rest
UndercoverToad: that's basically how beam search works
Pick8chu: AntiSquid thanks!
Cjx_1: Am starting to think like a sprite, I can see gold in sight! Its gonna be a good day :D
Pick8chu: lvl 50? that's crazy
Pick8chu: UndercoverToad okay thanks, i think i got what you meant a little
Pick8chu: I'll have to measure the state not actions
UndercoverToad: exactly
AntiSquid: that toad has a level 55 account, who's crazy now?
Pick8chu: wait what?
eulerscheZahl: hi
Pick8chu: that's just crazy lol
Pick8chu: hey
Pick8chu: wait
Pick8chu: that's the real one
UndercoverToad: yes
UndercoverToad: chat gets laggy with my main account
Pick8chu: that's why you are undercover toad!!!
Pick8chu: LMAO
UndercoverToad: i'm using the other one for the contest duration
Cjx_1: out of curiousity (im running a pure heuristic bot) what do the states map to, integers or rationals?
jacek: having frog avatar in french site :thinking:
UndercoverToad: up to you, I have doubles
Cjx_1: ^^ lol L'absurditie
UndercoverToad: and chromium just asked me if i want to wait or kill the other tab
AntiSquid: you think they'll stare at his legs jacek?
Cjx_1: good will to live in that one
UndercoverToad: they can eat daiver if they want
Cjx_1: rofl
struct: hi
UndercoverToad: hi legendary struct
struct: you didnt know euler?
UndercoverToad: i did
struct: ah
struct: thanks
UndercoverToad: this will be your cognomen from now on
UndercoverToad: until the next contest
struct: lol
struct: all it took was starting the contest 14 days earlier
UndercoverToad: you can do that again if thibaud needs testers
struct: probably wont do it again
UndercoverToad: competing is more fun
struct: yeah
struct: on first days I had no motivation at all
UndercoverToad: at least you did some testing, not like some other testers
BrunoFelthes: how many testes cg have?
struct: 4 + cg staff
UndercoverToad: how many testers did submit?
struct: all submited
BrunoFelthes: i was a tester for 1 contest... I did a good code, but after this they never call me again
struct: but a good ai?
struct: 0
struct: mine had 100% win rate
UndercoverToad: which contest did you test bruno?
struct: and i had a bot that had 99% win rate vs that one
BrunoFelthes: the one with baracks...
UndercoverToad: code royale
BrunoFelthes: this one
UndercoverToad: so the other testers did the equivalent of getting stuck in wood?
struct: all are > 2000
struct: ok above rank 2000
Noyotens: my rank is 1659 TH
struct: The names are on the forum I think
Noyotens: THIS IS GOOD RANK ??
UndercoverToad: depends on your goals
Noyotens: MY GAL IS TO BE WITH FIRST 1000
Cjx_1: THEN, WE GOTTA KEEP GRINDING HARDER nOY :d <3
Cjx_1: sry
UndercoverToad: CAPSLOCK IS ALWAYS GREAT
AlMualem: you need to be heard!!!!!
Cjx_1: when i loose in an fps its normally because shift key != caps key lol
UndercoverToad: reconfigure the key in your system settings
Cjx_1: double shift? Provocative.... Powerful....
iMobile: is it possible to retry a specific clash
UndercoverToad: https://eulerschezahl.herokuapp.com/codingame/puzzles/
Noyotens: I WANT LEARN AI ANY TuTorial
iMobile: ty
Cjx_1: iMobile, you can recreate the seed using added parameters at least. Not sure if you could get a specific opponents bot, short of coding their moves to be recreated.
Others would know better though.
Cjx_1: Ooooh :8
Cjx_1: thats awesome
AlMualem: guys i want to say ,i am really happy for the developers community...there is so much sharing :D
UndercoverToad: only during contests
UndercoverToad: next week the chat will be less active again
AlMualem: yes but there are forums and sites with questions and answers
UndercoverToad: and tons of helpful writeups for previous contests
michaelzlatkin1: Guys I'm so lost as to what you're supposed to do to prevent losing because a bot bumped you out of the way. Do you write avoidance? But wouldn't that just make you slower than if you just take the hit?
MSmits: are you talking about coders strikes back michaelzlatkin1?
MSmits: most players are currently doing a different game on this website
michaelzlatkin1: Yeah, sorry didn't realize this was a global chat!
MSmits: there is no other chat really, unless you're French or russian
michaelzlatkin1: Alas, I'm not
MSmits: it's ok to ask here
struct: euler you dont predict enemy right?
michaelzlatkin1: Thanks :)
MSmits: so anyways, if you get to the point where collisions matter, you need to start thinkign about a real simulation and search in your bot
MSmits: simple heuristics won't be enough then
MSmits: there's some sources on the csb main page
MSmits: but if you're looking to learn how to write a search/sim, csb might not be the simplest game to do this for the first time
elderlybeginner: to start brutaltester, do I need cg-brutaltester or I can go directly to the referee fork and clone the repo?
MSmits: I generally recommend playing this game until you get to this point, then try a different game, like a board game, to practice search/sim
MSmits: can always go back later
michaelzlatkin1: I see, yeah it's a bit unfortunate as I'm pretty junior and a company asked me to do this challenge as a code test
MSmits: I see, what is your current rank?
michaelzlatkin1: AI is really not my strong suit so it's been a challenge haha
michaelzlatkin1: Bronze 1
StevenV: haha bot programming for coders test
michaelzlatkin1: Or 3
MSmits: ah, let's take this to pm
michaelzlatkin1: Sounds good
MSmits: I'll help
michaelzlatkin1: Thank you so much!
BrunoFelthes: Anyone know a tool to merge java files into one?
NickSaidSomethingWonderful: https://github.com/Savinvadim1312/Java-File-Merger
BrunoFelthes: tks
NickSaidSomethingWonderful: I am using this one
Hmills: am i going to fail my gcse tomorrow ?
BrunoFelthes: is it works withut intelij?
StevenV: how to beat miyaza bot?
NickSaidSomethingWonderful: it uses a plugin in Intellij sothat it auto merges the files, without intellij, i am not sure
Sanguigno: ca
elderlybeginner: I'm looking for some help with brutaltester. Anyone willing to help?
DaNinja: ask away
Donotalo: how company rank is calculated?
aCat: score of top 5 players
aCat: sum of individual scores
elderlybeginner: I cloned: https://github.com/LSmith-Zenoscave/SpringChallenge2021
mvn pakage
got: INFO] BUILD SUCCESS
with some warnings
struct: feel free to hire me to climb
Donotalo: sum of top 5 players from the company?
DomiKo: yes
Donotalo: thanks
elderlybeginner: trying:
java -jar original-spring-2021-1.0-SNAPSHOT.jar -r "java -jar spring-2021-1.0-SNAPSHOT.jar" -p1 "python sc2021.py" -p2 "python sc2021.py" -t 2 -n 5
or
java -jar spring-2021-1.0-SNAPSHOT.jar -p1 "python sc2021.py" -p2 "python sc2021.py"
DaNinja: run brutaltester like this:
DaNinja: java -jar cg-brutaltester-1.0.0.jar -r "java -jar -Dleague.level=3 -Dlog4j.configurationFile=log4j2.xml ./target/spring-2021-1.0-SNAPSHOT.jar" -p1 "../x64/Release/Spring2021-ai1.exe" -p2 "../x64/Release/Spring2021.exe" -t 2 -n 30 -l "./logs/"
elderlybeginner: and getting respectively:
https://imgur.com/vt7Eq5w.png
https://imgur.com/Gi2oknI.png
Arces: ok
Chainman: Why y'all running brutaltester?
elderlybeginner: cg-brutal can be replaced with original-spring from referee clone or should I clone https://github.com/dreignier/cg-brutaltester as well?
DaNinja: you need brutaltester to test spring
Pick8chu: Sigh
ddreams: elderlybeginner for the second error message, you have too new java
ddreams: install java8
AlMualem: sum of top 3 players with >=5 participants
elderlybeginner: i'm getting errors, but it seems to work
openjdk 11.0.11 2021-04-20
elderlybeginner: https://imgur.com/ttfHkdE.png
seems to be legit
thank you for your help
Chainman: Hello
jrke: damn today I was in wood 1 of OOC someone pushed me in bronze and same code is now 1st in bronze lol
AlMualem: :D
AlMualem: gl mate
Marchete: no elderlybeginner
Marchete: use Java8
Marchete: in linux you can install as alternative
solistice: looking to implement prio queue in c#.. but that's slow, no?
Nerchio: its fine
UndercoverToad: too bad there is none in the default C# library
solistice: haven't found one :(
UndercoverToad: give me a moment, i have one that i copied from elsewhere
BoBot: is there a (relatively) simple function for (cellindex1, cellindex2) => areOnTheSameDiagonal?
UndercoverToad: http://chat.codingame.com/pastebin/e074ca3c-33a7-48aa-ae6a-1e865a138fe1
solistice: thx!
kovi: wow, nice euler
BoBot: I am looking at the numbers and I don't see it (and yea, I know I am about a week late with this :D )
UndercoverToad: thanks kovi. got a bit of code working finally
Cjx_1: bobot, consider what parameters are baked into the board. Maybe the cell neighbors can help you
kovi: you were beating me already last days, and now you also beat others :)
UndercoverToad: equal rights for all
UndercoverToad: except terry, that one hurts
elderlybeginner: Marchete I will.
BTW, what's not computed here?
WARNING: sun.reflect.Reflection.getCallerClass is not supported. This will impact performance.
BoBot: @Cjx_1 thanks, I will try some more - I know how to do it with "normal" coordinates and neighbor "deltas", but there are large jumps here :D
BoBot: oh, ffs, they are giving us the neighbors, nvm, I really should have started sooner...I am also trying out a new language
alchemsti: So, I'm suddenly getting a bad_alloc error in C++, which suggests to me running out of memory -- but tonnes of memory i s available (768MB I believe) and I'm pretty sure I'm not using that
alchemsti: (of course probably I am)
alchemsti: But -- has anyone else seen this? Coudl they have reduced limits?
alchemsti: I'm seeing it at early setup,where I really haven't allocated much of anything...
Marchete: incompatible Java versions, so better avoid any ERROR by using Java8. I had exactly that problem and changed to openjdk 8 while maintaining JDK11
Marchete: no errors so far
elderlybeginner: what do you brutaltester you bot against? Your other bot?
Marchete: me? crying with that many bugs
alchemsti: Hmm, I wonder if I have something too big on the stack...?
Marchete: for others is to change parameters and keep the better
elderlybeginner: so it's brutaltesting against your bots with different parameters. No access to boss thou
Marchete: yep
elderlybeginner: and you finish winning with your previous bot with 100% rate and loosing with everybody else :joy:
Marchete: inbreeding
Marchete: and it happens a lot
Marchete: it's like two children playing basketball
Marchete: and one of them always win
Marchete: and he try to play against NBA
Marchete: tries*
UndercoverToad: my 58% offline winrate translated pretty well
ClockSort: wtfffffffffffffff my bot has been predicting shade wrong this entire time i think
ClockSort: still looking into it...
Marchete: I have so many bugs I'll need another week
Marchete: wow euler
Marchete: congrats
UndercoverToad: thanks :)
Marchete: new feature?
UndercoverToad: yes
Marchete: great
Marchete: you still have new magic tricks after a week
Marchete: if you are topN will you write a PM?
Marchete: I always like these
Marchete: N is decided by you :D
StevenV: it is end of challenge now
struct: wow top 5
struct: grats
UndercoverToad: we'll see. there isn't much interesting in my bot even. but i have off tomorrow
StevenV: tomorrow will be ended
kovi: poor illedan
UndercoverToad: and I want to win Germany at least. that Philipp is strong
StevenV: still don't know how to go to legend
JohnCM: oh so exciting, i'm only 5-10 pts diff from the gold boss
MSmits: only?
StevenV: ?
JohnCM: ?
StevenV: :?
MSmits: or do you mean ranks?
JohnCM: nono i mean points as in game points
MSmits: 5-10 is a lot
MSmits: < 2 is a little
JohnCM: well this morning it was 60
MSmits: oh ok
MSmits: well good job then
JohnCM: yep just plain old heuristics
JohnCM: although i coded quite an elaborate one-ply search for growing
Chainman: really?
Chainman: I'm stuck in bottom of gold with heuristics.
JohnCM: yea prioritize growth which leads to a good outcome
StevenV: I keep getting this error
JohnCM: like gain more sun overall for yourself and blocks off opponent
StevenV:
Oups
An error occurred (#407): "You reached the limit of plays for a period of time.".
Sanguigno: try again in an hour
StevenV: :persevere:
JohnCM: yea i get the same error
JohnCM: i guess thre is a limit to the number of tries
StevenV: So I think this is why a lot of devs recommend brutal test
MSmits: yeah, it is much faster
StevenV: But I don't really know how to use it
Chainman: oh crap
JohnCM: i think the hardest part for my bot is that it doesn't adapt very well to the opponent fast strat
JohnCM: i take my time to grow trees, but if opp keeps harvesting, i lose a bunch of pts
MSmits: someone should really share how to do this halfway through the contest. If you look at chat history, you will find some tidbits about how to use it here and there
MSmits: forum post would be nicer though
StevenV: thanks
MSmits: can someone share the wiki chat history thingy
StevenV: I mean tomorrow is over, So I will try to use it later
JohnCM: MSmits, how're you doing for this challenge?
Chainman: I also didn't have time to figure out brutaltester.
JohnCM: have you broke into legend alr?
Chainman: I need to focus just on coding.
MSmits: JohnCM I am in legend yeah. I dont think i'll be going up much higher though. Hope for a place halfway through legend
MSmits: trying one or two more things, then it's done
JohnCM: all the best!
MSmits: thanks
JohnCM: haha i'm thinking of monte carlo now, since my bot doesn't adapt well. haha maybe i should have started off with that
Chainman: good luck
StevenV: I still in gold, but really can't figure out how to be legend
MSmits: a lot of things work JohnCM, but whatever you do is going to have to be supported by heuristics and pruning and some evaluation
Chainman: you are doing not that bad in gold JohnCM. I'm behind you.
JohnCM: definitely, i have plenty of heuristics and pruning
Pick8chu: Man
JohnCM: haha thanks Chainman, let's see if i can hit legend after a few heuristic tweaks
StevenV: some of my friends use monte but they can't even win silver boss
Pick8chu: these few days I was working on the challenge and learned so much through this chat lol
Pick8chu: you guys are so smart
JohnCM: silver boss not that tough compared to gold
Chainman: my last submit seems to be worse lol, I'm still like 100 places lower in this submit.
JohnCM: the main thing is to ensure your pieces are in "knights move" so they don't block each other. and harvest once you have >4 trees
JohnCM: should be able to get gold with that
StevenV: I mean hard to use Monte for this challenge
StevenV: I am already in gold
Chainman: https://www.mathsisfun.com/games/knights-move-2.html lol
Chainman: yeah keep them distance 2 to have knights move
Chainman: except for that one diagonal
MSmits: Monte works fine StevenV, you just need to do it in Python
Chainman: actuall you just need to check if it is collinear with distance 2 for it to be knights move
Chainman: I mean it should not be collinear
StevenV: I will try after this challenge end
MSmits: I was kidding :P
StevenV: I don't think I have time to implement monte now
MSmits: it'd be Monte Python is all
JohnCM: haha
JohnCM: i get the joke
JohnCM: but really, monte implement in C++ or C# is better
JohnCM: you get more iterations
MSmits: yup
JohnCM: python is for lazy coders like me who want to just get gold with heuristics
MSmits: well c# wont do that much better than java really
StevenV: I use java, which have problems with performance
JohnCM: so it's C++ then
MSmits: or Rust :)
StevenV: It's 3 years since I last use C++ in school
StevenV: :D
StevenV: really don't remember much of it syntax
Marchete: no MSmits
Marchete: you are wrong
Marchete: C# >>> Java
MSmits: i'm sure it's faster, but isnt it similar order of magnitude
MSmits: I mean c++ was always 2-3 times faster than both of these
Marchete: sometimes is
Marchete: I recoded something to C#
Marchete: got same performance
Marchete: using AVX2 and such
MSmits: you mean from c++ to C#?
StevenV: C# same performance as C++ really?
Marchete: yeah
MSmits: well i suppose using the same intrinsics, where these operations are the bottleneck, it makes sense
struct: omg top 30
struct: I cant look
JohnCM: the containers are very efficient
struct: o.o
Marchete: https://raw.githubusercontent.com/marchete/Mokka/main/img/Test2a.JPG
Marchete: https://raw.githubusercontent.com/marchete/NetMokka/main/img/Test2a.JPG
Marchete: NN inference engine, doing MNIST
Marchete: 12us/sample
Chainman: legend struct?
TobiasA: what does the circle beside the number on the leaderboard mean?
Marchete: good job struct
dbdr: progress status TobiasA
TobiasA: of submit?
dbdr: yes
Marchete: euler is going strong
MSmits: lol I accidentally set my beamsearch to exit when day >= beamwidth and it performed ok anyways
MSmits: i guess time runs out
peerdb: how small is your beamwidth?
dbdr: how did you manage that?
MSmits: 1000 :P
peerdb: < 24 then
MSmits: was gonna use BEAM_DEPTH, but typed BEAM_WIDTH
peerdb: ah gotcha
dbdr: if day > 1000, you can exit, yes ;)
MSmits: this contest is just full of funny errors for me
dbdr: long term planning
MSmits: looooong term
StevenV: I use Java and really don't know to do simulation without timeout errors
MSmits: dont make new objects
MSmits: generate a big store of them in the first 1 second
struct: thanks
MSmits: and then just reuse them
Trottel-Nit: Two tricks: Create most objects in the first turn (no "new" durcing game turns, use your own pool-arrays). B) "Warmup" the JIT, run some game loops on the first turn
struct: I didnt think my change would be this big
StevenV: Thanks
Trottel-Nit: that is a B ) ... no fun in there ...
Marchete: what change?
dbdr: nice struct! where were you before?
struct: 66
struct: not massive improvement but ill take it :D
flyingferret: wowfirst time here...and ouch i'm rusty
StevenV: Do we have any gift if we rank high?
struct: top 20 gets shirt
struct: top 3 gets art
StevenV: art?
struct: a canvas
StevenV: picture?
KelvinAndHubbles: the timer in the top left jus randomly turn red for anyone?
Cjx_1: bout to explode
Cjx_1: run
CouscousAgha: we are now in the end game
Chainman: yep
JohnCM: ok i can't test in ide anymore
Chainman: It's getting exciting for the day.
JohnCM: time to go rest i guess haha
JohnCM: see how it goes tomorrow with my new untested submit
Chainman: yeah that was me yesterday, and then I slept for 10 hours
Chainman: and cleared up my issue in like 5 minutes when wake up.
JohnCM: i'm losing by 5 pts to this super mario dp guy
JohnCM: aginer.. haha so frustrating
Chainman: dynamic programming?
JohnCM: display pic
Chainman: oh
ClockSort: super mario dp... haha
ClockSort: dp[mario][luigi]
TheBatMan_TM: hey guys
Marchete: reCurs_e plays completely different
TheBatMan_TM: how do i run sudo in repl.it?
Marchete: he has more than 1 seed in the board
Marchete: and not the usual starting moves
Marchete: https://www.codingame.com/replay/556410456
Marchete: SEED 24 11?
Marchete: at frame 6
ClockSort: i'm going to legend
ClockSort: gg
JohnCM: nice one clocksort
Cjx_1: grats
ClockSort: i had a bug where my MCTS thought small trees could shade large ones.
ClockSort: i fixed it and now I've won 43 in a row, no losses, in gold division
JohnCM: i actually removed the knights move requirement and my code is performing better
JohnCM: wow MCTS really works wonders
struct: yeah marchete
struct: but it makes sense in the future
ClockSort: mcts is magic
JohnCM: how many moves did you lookahead till
JohnCM: for mcts
kovi: competition is heavy for 3rd place
ClockSort: i get about 10k rollouts from a new position on day 7
JohnCM: or just random rollouts till game end (which is a little too much)
ClockSort: but then day 8 i often predicted my opponent's move so i can build off that tree
ClockSort: yes it's rollouts to the end position
Chainman: you removed knight moves, crap
Chainman: I need to learn mcts after this contest
Chainman: not enough time in 16 hours.
JohnCM: haha i don't think i can do that kind of rollout in python
Marchete: simplify to rollout 6
JohnCM: but i did a 2-day lookahead for tree growing to see if i should grow in a particular spot
Marchete: waits
Chainman: crap I was going to try the knight moves thing.
Chainman: It's worse though?
post4k: Hey guys, this might be a weird question, but how do I double check if the chat im in, is related to the CoC I just did?
I cant seem to find the CoC id on the page
Pick8chu: gosh dang ittttttttttt
Pick8chu: oh. post4k sorry i'm not sure
KelvinAndHubbles: post4k, no quite way, you just have to guess
post4k: hehe
post4k: Ill just guess then :)
KelvinAndHubbles: just do every clash chat and hope people dont think your crazy
Pick8chu: I changed to cpp from JS and totally forgot divisionbetween int give you int
Pick8chu: .....
ClockSort: 65 matches complete in gold so far, 100% wins. o_O
Chainman: wow
Chainman: that's awesome
Pick8chu: that really is, good for you
Wontonimo: i'm taking credit for popularizing "knights move" thing here. Do it Chainman!
ClockSort: oh, you mean Chainman's Opening?
Wontonimo: Hey Chainman - this video on MCTS is great. He even gives exact calculations so you can make unit tests of it https://www.youtube.com/watch?v=UXW2yZndl7U
Chainman: lol
Chainman: I'm doing knight move
Chainman: but I noticed bossoot from silver was doing knight's move as well.
Chainman: I just didn't do it, but try in gold.
Chainman: crap, maybe I'll try to learn mcts, I'm like 965 in gold now :(
Wontonimo: JohnCM - my MCTS bot easily gets to depth 27 with only 3K rollouts
Marchete: 3k rollouts to end?
dbdr: that might be too deep and not wide enough, not enough exploration
rohrokz: https://www.codingame.com/clashofcode/clash/1755144530fa86c4f2f3b265568cb9f926d4e10
Scarfield: But less depth means one has to make an effort of making an eval function :p
Wontonimo: dbdr thanks, you are probably correct
dbdr: I suppose the depth only count the tree part. you can still use rollouts to avoid an eval
CouscousAgha: you got this ClockSort
ClockSort: i know lol, it's not close
ClockSort: did you hear about my bug?
Wontonimo: Marchete - yes, 3k rollouts to end on around day 4. Day 1 only 2k.
Wontonimo: i've got some performance issues. it should be around 10-15k
Marchete: linear MCTS?
Marchete: like p1 first sim then p2 sim
Chainman: 3 ms for a turn with just heuristics lol :(
Chainman: so far I look better than my original bot with just addition of knight's move
Wontonimo: 1 player mcts. i may work on DUCT in a few hours
Chainman: I've won 4/4 with knight's move against myself
Chainman: I think it's an improvement
Chainman: okay 6/7 I guess
Wontonimo: it is. my theory is that it encourages growth into your opponents area and is a proxy for avoiding your own shadows of that one tree
AntiSquid: damn, should have done this strategy from day one :/
Chainman: I'm debating to try mcts or just add some search algos.
AntiSquid: fastest rank improvement
Chainman: what AntiSquid?
AntiSquid: random heuristics
Chainman: what's that?
Wontonimo: return actions[ rand()%len(actions) ]
AntiSquid: something that can get you legend Chainman
SirLoxley: Darn :/ I wanted to debug my simulator ... tried to use the offline coding game tool, didn't work and deleted it ... and ended up deleting my simulator, too. Guess that's game over for me. Good luck to all who still compete
Pick8chu: When you guys run on brutaltester, do you fight with your own bot?
Chainman: without search algos?
ddreams: Pick8chu yes
UndercoverToad: yes Pick8chu
UndercoverToad: :D
Wontonimo: oh man , that 's doo bad. SirLoxley !
Pick8chu: I seeee
Pick8chu: ddreams yo!
Pick8chu: morning man
ddreams: or, an older version of the bot
Chainman: git SirLoxley
ddreams: morning, having fun? :)
Pick8chu: I made it work with c++ haha
struct: ok now im a bit more motivated to keep going
Wontonimo: several older versions - so as to not overfit just the last one
struct: lets see if I can get top 20
struct: But cant get a shirt :'(
Wontonimo: go struct go!
AntiSquid: because of what i said struct right ?
Pick8chu: and was working on it but jsut found out that I've been dividing int/int which only gave me 1 or 0
Chainman: lol random heuristics doesn't even make sense
SirLoxley: I am more of a mercurial guy ... anyway that would not have helped me.
AntiSquid: ok what is your best gold rank Chainman ?
Chainman: umm I was 965, but just submitted with knight's move.
Chainman: hopefully break 800 now or something idk
Wontonimo: i think he meant "a heuristic, not anyone one in particular that has been discussed here" Chainman
Scarfield: what is knights move? seeding in "triangles"?
SirLoxley: I still have the compiled binary and an earlier core file. Not sure there's a lot I can do
AntiSquid: so look at my rank and of couirse it doesn't make sense but i am correct Chainman
Chainman: Sure, but I think I need to know what you mean by random heuristics.
Wontonimo: from chess Scarfield, step 2 in line, then one to the side.
Chainman: Cause just using some random heuristic obviously wouldn't work, or choosing random actions.
Wontonimo: basically, not in a straight line from the seeding tree
Chainman: or I'd be doing better right now lol
AntiSquid: from 900 to 440 now :D
Scarfield: yea got it wonto ty :)
Chainman: Scarfield https://www.mathsisfun.com/games/knights-move-2.html
AntiSquid: 414
AllYourTrees: nice AntiSquid
AntiSquid: 397 lol
Chainman: wow
ddreams: if you're doing really badly, the best strategy is to do the opposite of what you think is good aka. the reverse self-psychology strategy
Marchete: the opposite of bad is worse
AntiSquid: anyways i do agree it doesn't make sense because obviously the game is balanced and favors skill, right right?
Wontonimo: hey AntiSquid, what was your rank before knight move ?
AntiSquid: i don't have knight move
Chainman: you mean me probably
Chainman: I am not doing much better on submit with knight's move, but it was beating my version 6/7 when I tested it.
Marchete: knight move is automatically executed if you eval some kind of shadow alignment
Chainman: I'm still 900 rank
Scarfield: queen move is better obviously, because queen better than knight :p
Chainman: but I'm only 20%
ClockSort: go for pawn move, you might promote
AntiSquid: if (!bfs_mask[seed_idx]) gp->score+=10;
else gp->score-=10; Wontonimo
Chainman: all reverse psychology, just play chess
Scarfield: lol clock
Wontonimo: bfs_mask ?
Wontonimo: brute force search mask ?
AntiSquid: don't think about it, just do it
ddreams: Best FriendS mask
Scarfield: breath first Seed
Wontonimo: oh, i'm stupid
Wontonimo: thanks
AntiSquid: damn i have some timeouts for games i should have won :'( https://www.codingame.com/share-replay/556845158
ddreams: AntiSquid I'm getting an error about bfs_mask not defined.. help?
Chainman: lolololololol
Chainman: I have same problem
ClockSort: that's happening to me too
Wontonimo: how long are submits taking now?
struct: they are fast
Chainman: for me I'm 27% in like 5 minutes +
AntiSquid: 9/10 wow that's an amazing start @_@
ddreams: 13% in 10 minutes
struct: mine took around 1 hour?
AntiSquid: hope it keeps up!
Chainman: NameError: name 'bfs_mask' is not defined
CharmyRosewolf: So close to beating gold bot. I don't know if I can take this pain any longer :sob:
Chainman: I tried the random heuristics AntiSquid but I lose 39 - 151 to my bot lol
UndercoverToad: that's significant
Chainman: I do wonder if it is best to take like 5 best moves and just choose randomly
UndercoverToad: why not take the best?
dbdr: might be too good
PRO_G_r_AM_m_ER: will see what this does (hoarding ftw) https://www.codingame.com/replay/556865650
ddreams: surprise the opponent
Chainman: lol, nerf the bot, he is too strong
UndercoverToad: right, we don't need a 2nd recurs e. random of top5 is allowed
ddreams: maybe that's his strat
UndercoverToad: not nerfing it? that's not fair
ClockSort: ok i made it into legend
Scarfield: nice!
ClockSort: and i'm still rocking it :scream:
ClockSort: #34 and climbing
UndercoverToad: don't worry, reality will hit you when you fight #1
UndercoverToad: but well done for going that high after promotion
Chainman: whoah congrats ClockSort
ClockSort: i'm weak against bots who stockpile suns
ddreams: nice, cs
ClockSort: @UndercoverToad i had a bug where i let small trees shade big ones. I was top of gold with that bug (!)
ClockSort: fixing that bug launched me, um, very high.
UndercoverToad: that's a horrible bug :D
ClockSort: i know, it's crazy that i won any matches
Chainman: wonderful bug for others though.
struct: its still mcts ClockSort?
ClockSort: @struct yep, duct mcts
kovi: gratz clock, duct has a chance for top20
Chainman: http://mlanctot.info/files/papers/sm-tron-bnaic2013.pdf
Chainman: is this duct mcts?
DomiKo: 4.2 yes
UndercoverToad: 4.2 what?
Chainman: section
UndercoverToad: february the 4th?
UndercoverToad: aah, sorry
Smelty: nah, april the second
Smelty: oop
UndercoverToad: somehow I filter links while reading
Scarfield: 1/100 of a good time
ClockSort: i invented duct and then discovered it had a name and then found that paper :P
UndercoverToad: smits needed 2 years to find a paper about it
Scarfield: duct is smitsi?
UndercoverToad: it's not? i didn't read all the details
Scarfield: no idea, havent looked into it
RoboStac: no - smitsi has separate trees per player, duct is one tree
Scarfield: saw da_po call it UCT forest
dbdr: I think there's a paper for what MSmits does too, isn't there
Smelty: wait w0t
MSmits: I'm not sure
RoboStac: yeah, theres a few very similar things. Can't remember the names though
Smelty: yes
UndercoverToad: oh, we summoned him
MSmits: :P
Smelty: h-him?
UndercoverToad: smits is a he
MSmits: did not get around to trying smitsimax in this contest btew
Smelty: he?
Smelty: hmm
yes
UndercoverToad: you told me to try it but I never intended to give it a try :(
UndercoverToad: should I feel guilty now?
MSmits: nope
Scarfield: yes!
MSmits: I am not sure if it is worth trying
MSmits: this game has me unsure about everything
CogitoErgoProxy: hi guys i've not undestood how do i recognize a undiagnosed sample from a diagnosed in code4life, do you have hint?
MSmits: err... that's pretty random
MSmits: I understand csb questions, but...
UndercoverToad: score -1 i think
struct: I think its time for an all nighter :D
MSmits: not me, i am quitting in an hour or two i think
MSmits: maybe i'll do some netflix 'n brutaltest (no chill)
struct: maybe ill do some aws and wait
UndercoverToad: maybe my computer has to sleep in the bath tonight
Chainman: hahahaha
Chainman: alright I'm rank in top 800 now in gold. I'm ready for searching.
MSmits: mmh I just wrote opponent prediction beamsearch, integrated into my own beamsearch and I get the same wr.... thats just annoying
MSmits: i mean, if it was much worse... ok, but 50%, wth
UndercoverToad: congrats, for me it got worse
Chainman: how do deal with score when searching moves ahead?
UndercoverToad: yes
Chainman: I'm not sure, do I just add score for a path, and take path with greatest score
Chainman: instead of just choosing next move with greatest score.
Cruiser18: Is there a way to try some of the Clash Of Code challenges, without actually being pitted against other people?
MSmits: depends on your search Chainman
MSmits: i dont know what you are doing
MSmits: is this a beamsearch?
Chainman: I'm going to be doing a beginner simple search algorithm, maybe prune it to be like beam_search
MSmits: then i dont know what to tell you except yeah you somehow take the best path/highest score
MSmits: searches are all different
Chainman: I think I'm going to be doing beam search lol
MSmits: with that one you take the highest scoring state at the last finished depth level and pick the first move of its path
Chainman: I mean I tried earlier, but gave up on it, and now I have better heuristics.
MSmits: so make sure you can reconstruct the path or at least keep a reference to the move done
Chainman: I had made it too complicated and it was super slow.
Chainman: I see thanks MSmits
MSmits: np
AntiSquid: #366 Chainman
Chainman: nice, still not sure what you did lol
azriel1rf: Now, I'm in Legend!
azriel1rf: in Spring Challenge
AntiSquid: are you looking for a job or you wrote that for fun in your profile azriel1rf ?
azriel1rf: i have no job,,
azriel1rf: is it prohibited to write such thing?
ddreams: no
azriel1rf: ok
AntiSquid: no was about to ask if that's why you were committed to get legend :P
azriel1rf: lol
MSmits: could change it to: *WILL WORK FOR $$$"
azriel1rf: you said it
azriel1rf: i use much time to
azriel1rf: this challenge
aangairbender: wow top1 is so far ahead
MSmits: yep, mostly unbeatable
aangairbender: neural network?
AntiSquid: go NN or go home
MSmits: it is my guess yes
Westicles: opening book+SA+closing book
BrunoFelthes: i won 1 against reCurse, then i realize that he timeout
AntiSquid: no, we are past that question . which algo is it is the question
azriel1rf: learn locally and implement it into code?
MSmits: share with us how to make him timeout :P
LoganWlv: Haha
Kellthazar: Jeez...
Michael_Howard: I'm losing with "-" score in Arena games I win when send game parameters to IDE - is that definitely timeouts?
AntiSquid: Scarfield hi how's it going, what did you do for your bot? :P
ddreams: or crashes
ddreams: or memory limit exceeded
AntiSquid: MSmits beam or mcts ?
Michael_Howard: I don't have a tree :)
ClockSort: @Michael_Howard I am getting the same results
MSmits: beam AntiSquid
Scarfield: hi squiddy, "guided" MC both players UCT for first action
ClockSort: sometimes '-' in battles
Westicles: Is dapo a NN guy?
MSmits: might have tried a different one if i did not spend so much time fixing dumb bugs
Scarfield: rest actions random
Scarfield: sim till end of game
MSmits: Westicles doubt it, he won that sponsored contest without NN
AntiSquid: Westicles why are you asking about him ?
BrunoFelthes: i'm using GA
Westicles: Just wondering if anyone can swoop in and win
MSmits: the only way to do that is to somehow find out what makes re curses bot fail and keep it hidden until the last moment
MSmits: not the way i would want to win though
AntiSquid: probably if the contest prizes were as big as for the private one Westicles
ddreams: lol, did a change that looked good to me just testing losing games without testing it in brutal
ddreams: much better arena rank
ddreams: 36% wins in brutaltester
MSmits: arena is pretty random though
ClockSort: is grow 33 a mistake on round 4 in this game against Mazelcop? https://www.codingame.com/share-replay/556895891
BrunoFelthes: what is the prizes for private contests AntiSquid?
MSmits: check how much your rank increase corresponds with in terms of rating ddreams
MSmits: if it's less than a point, it could just be random
BlitzProg: MCTS being built... looks hard to stabilize
BlitzProg: so far, 0.4ms per rollout but crashes eventually
solistice: someone please kill me
UndercoverToad: it's certainly not optimal ClockSort. but i'm not sure if my bot would even do the same bad action
solistice: been working for hours fixing something and not figuring out why I coulnd't reach gold
UndercoverToad: do you have the input for that turn? i'd want to know what I would do
kovi: talking about surprise - RoboStac is also capable of doing mcts-nn
solistice: saw that I stopped after 1 iteration of Iterative Deepening for testing purposes
UndercoverToad: 1 or 2 days ago he said he's stuck
solistice: so was consantly looking depth 2
ddreams: MSmits is there a way to check what rank I had?
MSmits: no, but i would guess you knew, because you said it was doing better?
ddreams: well, I never broke top 100 before
UndercoverToad: and he's not the guy for last minute surprises. was leading code royale from start to finish
kovi: true
MSmits: robo does not hide I think
JBM: who does though
BlitzProg: nvm, wasn't reseting action count when rolling, surprised it wasn't crashing earlier lol
BlitzProg: it's... somehow better than my MC so far? that's insane
kovi: i didnt meant hiding, just getting nn ready over a pure/nonheuristic mcts/duct
ClockSort: @UndercoverToad i don't know how to get the input
ClockSort: the reason i play grow33 here is because i'm favoring large trees in a heuristic. https://www.codingame.com/replay/556895891
UndercoverToad: i always print stdin directly to stderr
ddreams: Illedan still improving your bot, or do I have a shot at #1 in the country? :)
ClockSort: i think i can just remove that heuristic and rely on mcts
UndercoverToad: then i can copy-paste from the replay info
UndercoverToad: Illedan promised no sleep until he beats me
struct: lol
Chainman: let's go I'm 670 in gold
ClockSort: i am going to try removing all opening heuristics and see if my less-buggy-mcts can outthink them
ddreams: clearly inspired by Harald Hårfagre.. the king who refused to cut his hair until Norway was one kingdom
AntiSquid: is this stuck at 300 ? annoying @_@
Scarfield: i caught up to you againg though, whats your approach squid?
AntiSquid: Chokudai no kensaku arugorizumu
UndercoverToad: new keyboard layout?
ClockSort: ok, i removed my opening, now mcts handles that too.
azriel1rf: chokudai search? (beam search with efficiency to time management)
Scarfield: according to google translate, Chokudai no kensaku arugorizumu = A little sickness
struct: Im using chokudai
struct: I think
azriel1rf: chokudai no kensaku arugorizumu = chokudai's search algorithm
struct: thats the translation?
azriel1rf: yes
AntiSquid: oh it's supposwed to be time efficient? when the width expands too much my bot times out
Scarfield: not according to google translate :p
struct: http://chat.codingame.com/pastebin/e4493ade-9281-421a-9a4b-429c02f8c180
azriel1rf: no = の = 's
kensaku = 検索 = search
azriel1rf: arugorizumu = アルゴリズム = algorithm
ddreams: you may be Japanese, but Scarfield trusts google
AntiSquid: Scarfield look at his location, who do you trust? google or the native speaker?
Scarfield: i believe you, just found it funny it was translated into "a little sickness"
ddreams: well.. when I ask google voice translate to translate "it's martina's birthday and she's very happy" it thinks I'm saying "it's my pen*s birthday and she's very happy"
AntiSquid: you use a vector for it struct or what ?
struct: yes
Scarfield: thats because it always listens, and translates what you are most likely saying :p
AntiSquid: tried anything else ? linked list or somesuch ?
ddreams: I was trying to demonstrate that her mother and I could use google translate to communicate (she's italian)
ddreams: lol
Scarfield: xD
struct: vector is fast enough
struct: I barely use it
struct: I use it like 10 times per turn at most
AntiSquid: maybe i implemented something wrong
struct: maybe I also did :p
Chainman: so basically I kind of have to simulate gamestate in my beam_search, only thing that makes sense.
Zeph.: .
Chainman: I want to beat gold boss at least once.
Scarfield: thats the point of "searching" Chainman. you look some moves/actions ahead, evaluate the state, pick the move which leads to the most promising state
Chainman: yeah just checking it's my first time.
cegprakash: Chainman just mimic gold boss
Chainman: wasn't sure I was doing too much/unecessary work
Scarfield: important node, if you havent thought about it, assume your opponent is trying to get the best score as well
Scarfield: node = note
Chainman: I like to assume my opponent is bad though :stuck_out_tongue_winking_eye:
Chainman: Wow recurse is far in lead.
UndercoverToad: he's no fun playing against
DomiKo: top20 is hard now :/
UndercoverToad: fix it
DomiKo: i'm trying :sob:
UndercoverToad: 408 in submit
DomiKo: I can't add any feature to eval that don't break my bot :9
DomiKo: :(
UndercoverToad: been there, dumped eval and started at 0
DomiKo: ehhhh
struct: I think the only way I can climb now is if I had any enemy prediction
struct: I still have 0
ddreams: 408 in submit? is it broken again?
UndercoverToad: top 20 is possible without prediction
struct: :'(
jrke: i won't be in legend another contest again :disappointed:
ddreams: I'm at 73% after two hours :<
UndercoverToad: 56%, 1h 17
jrke: gn
ddreams: perhaps ping thib already?
UndercoverToad: not sure if system problem or sunday business
ClockSort: not sure if sunday problem or system business
UndercoverToad: i have an echo
Alexiada: y
Alexiada: h
ddreams: echo off
Alexiada: salut
dbf: good that I don't need to submit now :) can see that it is not better from 10 games :(
ddreams: just passed Illedan.. I hope he's sleeping already
UndercoverToad: it's 7:43pm for him
Krakenus00: Wow, you were prepared
ddreams: and me
UndercoverToad: and he's online on discord
ddreams: here too
UndercoverToad: i didn't realize you are from Norway too :D
struct: ping him :)
UndercoverToad: thibaud?
ddreams: nooo, let him sleep
struct: Illedan
dbf: or cut his internet cable
struct: ok 14 hours left
UndercoverToad: time to start coding
struct: yes
ddreams: time to completely reimplement everything
struct: its time for clash
struct: I will rush now
struct: my brain is fresh
ddreams: try the escape room?
struct: already did them
UndercoverToad: i did half for a room
Zenoscave: time to fix my eval issues
UndercoverToad: half of*
struct: You didnt like it euler?
UndercoverToad: meh
struct: yeah its not for everyone
struct: I enjoyed it tbf
ddreams: I love real escape rooms
UndercoverToad: i want clear tasks, no riddles where I have to find out what to do
ddreams: zeno, you're throwing away points in the last day
AntiSquid: you didn't see the worst part of it UndercoverToad
Zenoscave: does anyone know who in top 20 uses Beam Search?
struct: im top 30
struct: 31 :(
Zenoscave: with beam?
UndercoverToad: i know that I use beam
kovi: me2
struct: chokudai
casmith789: waiting for all the games played is nailbiting stuff
Zenoscave: how many sims (turn sims not day sims) do you guys finish?
struct: 1.5mill +-
Zenoscave: welp...
Zenoscave: 50K isn't gonna help then
peerdb: 1.5m is insane
ddreams: depends on your eval zeno
ddreams: 50k is sufficient
ClockSort: 1.5m sims??!
Eglerion: I should go switch language. Only at about 2k in PHP :(
Zenoscave: It's absolute trash
UndercoverToad: 40k for me :(
ddreams: 200k for me
ClockSort: in fall 2020 i was getting 150k and it was pretty lean. hard to imagine 1.5 million. are you storing them in an array?
DomiKo: a lot of seeds I guess
UndercoverToad: them = nodes?
ClockSort: yes, the object that contains, for example, the height map
struct: What do you mean array?
Zenoscave: I only allow one seed at a time. Is that not a valid pruning?
struct: I store the states on an array yes for bs
UndercoverToad: big fat static array
Marchete: struct 1.5mill?
Marchete: that's a lot
struct: yes
struct: GameState* Agent::states[3] = {new GameState[1000000], new GameState[1000000], new GameState[1000000]};
struct: GameState* currentStates = &states[0][0];
GameState* nextStates = &states[1][0];
GameState* finishedStates = &states[2][0];
ClockSort: yeah, I use this:
static State STATES[MAX_STATES];
therealbeef: 4.5M in first turn, 500k in regular turns
ClockSort: well i use a vector and alloc on Init, so it goes on the heap
Zenoscave: fix it ClockSort
ClockSort: @Zenoscave what do you mean, what's to fix?
UndercoverToad: first turn will be WAIT, using more time for that sounds useless
struct: my state is 96 bytes
AntiSquid: why not prune some of them lol
ClockSort: pruning takes time
ddreams: UndercoverToad even so, it's nice to have the number
Zenoscave: ClockSort stack tends to be faster here, iirc
peerdb: damn, 96, my state is double that
AntiSquid: anyone successfully transfering the search tree from turn to turn?#
struct: What do you mean by transfering?
UndercoverToad: 136 bytes per state for me
ddreams: rerooting
AntiSquid: first turn i get 330k but that's useless :D
ClockSort: my state is huge, i don't even know how big. it has an array of 37 ints for each height. not the best, ha
ddreams: I get 2.5m first turn
AntiSquid: so what do you do with it afterwards?
ddreams: throw it away of course
struct: array of 37 ints?
Zenoscave: 288 bytes here
struct: oh god
ClockSort: yeah
ddreams: that's how I started too
ClockSort: how do you store heights and check shadows?
ddreams: now I have four uint64_t's
struct: bitboard
ddreams: bitboarding
ClockSort: any reference/link for details?
Zenoscave: struct do you not store forward days of shadows?
AntiSquid: bitset
struct: nope
Zenoscave: trees[player][size] |= 1ul << cellIndex
struct: https://core.ac.uk/download/pdf/33500946.pdf
ddreams: do you cache sun/shadow computations?
ddreams: ?
struct: nope
ddreams: oops, double q
struct: my shadowing takes 10 nanoseconds
struct: so idc
UndercoverToad: what's that zeno?
ddreams: the bitboarding pdf by cameron browne
ClockSort: thank you struct for the link
struct: n+
struct: np*+
jacek: np-complete+
Zenoscave: for clocksort that's a basic bitboard encoding
Scarfield: https://eprints.qut.edu.au/85005/1/__staffhome.qut.edu.au_staffgroupm%24_meaton_Desktop_bits-7.pdf ClockSort
struct: I could share my bitboard structure
struct: is not amazing but I guess I could
struct: after it ends
struct: I think more people use it anyways
Scarfield: dammit you won struct, even a cleaner link
ClockSort: mcts was so hard to code (my first time) that i haven't spent much focus on making the state lean
ddreams: well, you did beautifully apparently
UndercoverToad: MCTS gets significantly better with more sims
Scarfield: after contest i can recommend clobber for playing with bitboards, probably the simplest game to do that for, or maybe breakthrough
ClockSort: very true @UndercoverToad
struct: I still have time for mcts
struct: but I have no idea how to implement duct
ddreams: how many rollouts did you say you get, clocksort?
ClockSort: roughly 10k from day 4, luckily i often predict the opponent's move and can reuse the tree on the following turn
struct: 10k rollouts is how many sims?
ddreams: 250k?
ClockSort: my rollout is not a full sim. i cheat a lot, for example, don't check seed distances when seeding during rollout
ClockSort: i just let them seed anywhere, it seems to work :grin:
Scarfield: :scream:
UndercoverToad: surprising
struct: o.o
ddreams: I've heard about such shortcuts, but didn't think it'd work here
ClockSort: after midgame the whole board is reachable anyway
kurnevsky: I had ~1500 sims in my mcts :)
struct: ok im gonna implement it
ClockSort: @kurnevsky me too, the O3 optimization helped
ddreams: duct too, kurnevsky?
struct: I have it coded already anyways
kurnevsky: But probably I did something wrong - seems too small :)
ddreams: I coded it up for fall.. perhaps I should try too
kurnevsky: No, I don't use mcts because of this :)
ddreams: aha.. beam search?
kurnevsky: Plain minimax :)
ddreams: nice, didn't think it'd work here
UndercoverToad: even a non-parallel algo where the opponent can move after you?
kurnevsky: Yes.
UndercoverToad: and that huge branching
AntiSquid: read chat history (wiki page) i said anything would work :P
casmith789: finally legend :)
UndercoverToad: i discarded the idea on the first evening already
AntiSquid: gz casmith789
casmith789: ty
UndercoverToad: dbdr still owes us some days of chat history
kurnevsky: I had parallel version but then realized that it actually executed on single core :)
UndercoverToad: and he should lock out the spammers that make the server lag
ddreams: wiki page?
dbdr: easy to say :P
UndercoverToad: with non-parallel I mean sequential (you take a step, then the opponent)
UndercoverToad: dbdr isn't there a setting that you have to approve profiles before they can edit?
Kellthazar: Im watching the alphago documentary to, maybe, have some idea to improve my heuristc XD
kovi: recurse submit
dbdr: yes, that's a good idea
emagnidoc: hello, i m not sure if this is the right place to ask questions but i m having a weird issue where my test cases are green but the test case during submit is not. thx in advance
UndercoverToad: while the game itself is concurrent, not sequential. which made me think that minimax would be too paranoid
ddreams: takes hours for submissions now
UndercoverToad: ping staff again or sit it out?
dbdr: it's a broken reCurse submit I think
struct: do it euler
ddreams: I think pinging staff is worth it
ddreams: costs me nothing
UndercoverToad: your turn this time struct
UndercoverToad: i already did on saturday, you can fully ruin the weekend
struct: lol
kurnevsky: > with non-parallel I mean sequential (you take a step, then the opponent)
Yes, like this :)
struct: ill just wait for some random to ping them
UndercoverToad: only mods can ping @staff
dbf: maybe they have on-call rota and there is a next person already :)
UndercoverToad: submit count is stable around 400
UndercoverToad: i still remember my 1st contest here (HyperSonic)
submits on the last day were extremely slow. like 10h
UndercoverToad: servers were crying in pain
Chainman: RIP
Zenoscave: ok 150k sims in 100ms now
UndercoverToad: still 40k :(
UndercoverToad: but heavy eval
Zenoscave: well that's my local number. I'm sure submit will be slower
Zenoscave: 60% timeouts :/
kurnevsky: Is it number of full game evaluations till the end? :)
UndercoverToad: 40k is my number of single actions I can simulate
kovi: 80k heavy eval
Scarfield: i wonder how expensive it is to increment the simcount
natofp: Why does it make sense to run random algorithm in MCTS instead of using heuristic to evaluate the position?
joelthelion: natofp your heuristic will bias your rollouts and that will make your MCTS miss possible counterplay
UndercoverToad: it's just hard to come up with a good eval. but adding some domain knowledge can still help if done right
joelthelion: I think slightly biasing your rollouts can work, but it's tricky business
ddreams: the number of seed actions vastly outnumber the rest, so
ddreams: I guess you can cheat
natofp: but how does it make sense in chess for example? most of the games would end in a draw so you should probably eval after some random rounds?
joelthelion: well, MCTS doesn't work too well for chess
kurnevsky: Biased eval should work, at least pachi uses it.
DomiKo: rebless recalc?
DomiKo: now 100% winrate is coming?
reCurse: Trying
Scarfield: looks much better than the prev resubmit :muscle:
UndercoverToad: i got a lucky win vs your last in arena but didn't save it
Scarfield: 100% or delete account?
kovi: i saw it...was thinking to finetune vs. recurse weakness for finals
VizGhar: so its NN right?
reCurse: :zipper_mouth:
kovi: but now im too weak for top3 anyway
Zenoscave: A Magician never reveals his secrets
Zenoscave: their*
UndercoverToad: tuning vs recurse gives you a good boost for each win
ddreams: you had some timeouts?
Zenoscave: if you can tune that is.
UndercoverToad: outweights at least 2 losses vs equally ranked bots
reCurse: I can't do anything about the timeouts
reCurse: Server hiccups are impossible to plan against
ddreams: just use 5% of your cpu to check the clock like me :p
reCurse: Just trying to fix its moments of stupidity
jacek: sidechannel attacks for you eval?
UndercoverToad: you mean the moments where it completely stops playing and loses by 50 points?
reCurse: Among other things
UndercoverToad: does it try to get a high score even? or just win/loss?
reCurse: :zipper_mouth:
UndercoverToad: ok
Astrobytes: nn confirmed :P
UndercoverToad: will you tell tomorrow?
PRO_G_r_AM_m_ER: ok, hope to finally beat silver, omg
reCurse: Sure, trying to stay focused
reCurse: Gonna close chat soon
UndercoverToad: ?disableChat
Zenoscave: ttyl reCurse GL on the 100%
struct: is your goal 100% winrate?
Astrobytes: It'll be a good PM, well done reCurse
reCurse: Minus hiccups yes
Scarfield: yea jebus, you are just being mean at this point :p
reCurse: Sorry
ddreams: sorry not sorry I believe
jacek: hes canadian, the sorry is legit
Scarfield: xD
Astrobytes: :rofl:
reCurse: That man canadians
peerdb: whats PM?
Scarfield: post mortem
peerdb: aah gotchu thanks
Scarfield: or private message in some cases, but not this time :)
ddreams: 100% completed battles now, up to rank 70, even if brutaltester gave 36% wins :o
Zenoscave: which I always thought was a weird nomeclature
Scarfield: ^
peerdb: yeah RIP recurse I guess
Astrobytes: Wonder if the Canadians got the 'sorry' thing from us Scots, we vastly overuse it
ddreams: number of concurrent submits have started to drop, ,nice
Zenoscave: dbdr https://www.codingame.com/share-replay/556973528
ddreams: maybe staff pinging helped
Astrobytes: don't think anyone pinged staff
Zenoscave: I lose 60% of the bottom league. you got a bug
Scarfield: stru ct did i think(?)
struct: nope
Zenoscave: oh Hai mark
ddreams: the toad did, I though
ddreams: t
dbdr: yes Zenoscave? :)
ddreams: maybe not
Astrobytes: Nothing on discord, unless it was in DMs
Zenoscave: you've got a bug. I shouldn't beable to even tie your bot
Zenoscave: Bad search perhaps?
dbdr: that's broad :D
Scarfield: fix it
UndercoverToad: i did not ping staff
ddreams: zeno's bot is optimized to beat dbdr
UndercoverToad: told struct that it's his turn today
Zenoscave: I don't know how to be less broad. just suprised at a win with a test bot that rarely wins
ddreams: ah, right, looks improved anyway
UndercoverToad: and submit count even below 400 again
ddreams: yeah, my last battles suddenly finished too
UndercoverToad: might be just a busy day. it always gets a little slower at the end
UndercoverToad: and start of the contest as well
Scarfield: temperature of server room :thinking:
Zenoscave: is a true skill elo of 9.3 bad ?
Scarfield: that depends
ddreams: val got a win against recurse... time to delete
reCurse: Happier about that loss than some others I saw before
UndercoverToad: 2 points difference and you were on the bottom side
ddreams: some maps are very biased too
UndercoverToad: which I consider at disadvantage
ryoukai: What time does CP update daily?
UndercoverToad: 3am CEST or something like that
UndercoverToad: which would be in 6-7h
Illedan: Hmm, ddreams above me. Ok, can't have this
ryoukai: Thanks
peerdb: Question: Im trying to write a beamsort and to prune my nodes I have to sort them on value somehow. std::sort seems very slow, what are good ways to prune gamestates
ddreams: bring it ;)
UndercoverToad: you can't have me about you either Illedan. do something about it ;)
Zenoscave: I hate having a toad about me
Illedan: I have 4 hours before bed. Wasn't allowed to be tierd tomorrow :D
Astrobytes: zeno :rofl:
UndercoverToad: a toad about you?
UndercoverToad: is that a pun that I don't get?
Astrobytes: UndercoverToad 07:39PM
you can't have me about you either
UndercoverToad: :(
Scarfield: xD
Zenoscave: :P
Astrobytes: :D
ddreams: toad doesn't even know what he's saying at this point
Scarfield: MansplainstroBytes
UndercoverToad: my fingers just don't obey me anymore
Zenoscave: I think it's sleeping time toady
Astrobytes: Cat petting time
Chainman: yeah
UndercoverToad: got a version which is 52% to 48% in local
UndercoverToad: no idea how that translates to the arena
SabertheLost: If the shadowsize is 2, then a tree size 3 get zero? Correct. But I´m looking at a board where blue prints get 17 but on board it should only get 12
ddreams: lol, I have no faith in brutaltesting myself anymore
ddreams: I'm happy I submitted this version without testingit
ddreams: gave me like 50 ranks
peerdb: what is this brutaltest youre talking about?
ddreams: even if brutal said 36% or so wins
UndercoverToad: play offline vs yourself, several games in parallel
ddreams: peerdb a program that pits your bot against itself many times
Chainman: SabertheLost you get sun points at the beginning of the day
akagami101: lol
Scarfield: trees only cast spooky shadows on trees of same size or smaller SabertheLost
SabertheLost: Chainman. Is that that before or afte the sun is moved?
ddreams: SabertheLost a size 2 tree does not spookyshadow a size 3
SabertheLost: Is that a change of the rules from the first days?
ddreams: no
Scarfield: no
ddreams: They will find the shadow on a cell spooky if any of the trees casting a shadow is of equal or greater size than the tree on that cell
UndercoverToad: still 200 players missing to hit the count of last contest
SabertheLost: ah crap another rule that hidden in text. Or was it explained different in the youtube viedo? I think that viedo might done more harm...
Scarfield: "They will find the shadow on a cell spooky if any of the trees casting a shadow is of equal or greater size than the tree on that cell. " last part of Sun & shadows
ddreams: the text really emphasizes the wrong things
Scarfield: ^
Scarfield: yea you have to read carefully
ddreams: Helping the wood spirits are LESSER SPIRITS hiding among all the trees.
They will find the shadow on a cell SPOOKY if any of the trees casting a shadow is of equal or greater size than the tree on that cell.
ddreams: the bolding is just misdirection
ddreams: In between each day, the sun MOVES to point towards the next direction
Scarfield: :o didnt even notice
Scarfield: it is hard to write a concise and easy to understand statement though
SabertheLost: That makes my whole algo bad.
Scarfield: good news, "easy fix" to gain in rank maybe ;)
SabertheLost: I think I was confused that the distance did not matter and assumed that distance ment also size
SabertheLost: and no my algo sucks. :)
Scarfield: xD
SabertheLost: It´s just number of trees shading an square
Scarfield: no no, not the number of trees, only their sizes matters
Chainman: alright, I'm getting 40 ms with a very simple beam search of width=1 and depth = 4 lol. Much better than last time. But I'm also not consider opponent, and not consider shadows yet.
Chainman: I need to count the number of nodes I can get this time.
ddreams: 4 days?
Chainman: nope
Chainman: 4 actions
Chainman: wait and my width was 1
Chainman: wtf, that's sad
ddreams: better compare nodes after wait
ddreams: *for move selection
InternalErrror: Why is the Python Interpreter here still not 3.8 or higher?
InternalErrror: Walrus would be great for Clash of Code shortest mode
struct: its 3.9.2 ?
InternalErrror: no, walrus operator is always marked as invalid syntax
InternalErrror: :=
struct: but does it compile?
InternalErrror: no, syntax error
UndercoverToad: can still work
UndercoverToad: highlighting is not the same version as the language itself
jacek: who needs walrus, sounds pascalic
struct: it compiles
InternalErrror: It's assignment + return the value in one operator
jacek: oh i know! CG needs new language, delphi!
UndercoverToad: that's Pascal in OOP
UndercoverToad: on paradigm that's really missing: Prolog
ZarthaxX: hate prolo
ZarthaxX: g
annaproxy: Yess prolog
UndercoverToad: never tried it
ZarthaxX: used it on uni
ZarthaxX: damn man
UndercoverToad: just saying: we have a lot of OOP, some functional
UndercoverToad: prolog is a thing for itself
struct: whats prolog used for?
CouscousAgha: can i debug print in the console output?
UndercoverToad: prolog is a logic language
InternalErrror: http://chat.codingame.com/pastebin/e8064795-fd40-4332-a538-e5b9e35dc69d
UndercoverToad: do you know the Einstein riddle?
struct: true ai?
struct: no
reCurse: Ew prolog
UndercoverToad: where different persons live in houses, have favorite food and nationality, houses have colors
UndercoverToad: you are given a bunch of relations
UndercoverToad: and have to find the actual assignments
UndercoverToad: for prolog you just give the relations and the program will solve it for you
UndercoverToad: https://www.codingame.com/training/hard/einsteins-riddle-solver
InternalErrror: what now it works
InternalErrror: just one clash ago it didn't
struct: cg staff is fast
struct: they just updated based on feedback
struct: ill check it afte contest euler
ddreams: updated what?
InternalErrror: insert * flash meme *
ddreams: python version?
InternalErrror: yeas
UndercoverToad: obviously struct is trolling
UndercoverToad: * it's sunday
- nothing here is done that fast
jacek: crashing is done fast
struct: ok i Need to fix my damn bot
ddreams: also changing python version in the just as a contest is finishing :D
struct: I said this 3 hours ago
ddreams: - in the
UndercoverToad: i think i'm making mine worse
UndercoverToad: or i was lucky before
ddreams: doing mcts struct?
struct: no
struct: beam
ddreams: thought you'd test it
struct: Maybe I will
UndercoverToad: top20 or delete
struct: But I dont know duct
struct: and I think i cant understand it
ddreams: I'm trying to get my fall mcts to work
struct: I dont feel like top 20 is realistic
ddreams: delete
struct: maybe on next contest
struct: im safe on this one
jacek: tomorrows contest ends and we can finally go back to clashes, the main point of CG
struct: ok its time for me to close the chat
struct: cya in 12 hours
ddreams: ciao
Donotalo: server becomes so slow during the last 24h of the contests, this is frustrating :(
Donotalo: it takes a lot of time before knowing my new bot's position
Donotalo: why don't CG do anything about it?
DomiKo: what they can do?
Donotalo: may be increase number of servers?
DomiKo: and who will pay for this?
Donotalo: just thinking, as it happens always
Donotalo: ha ha, yea, that's the question
skotz: maybe get some corporate sponsors
Donotalo: in last 4-6 hours, i hesitate to submit any new code because of this slowness
ddreams: I'd pay for it personally
ddreams: cost of joining a contest instead of working for clients is pretty high
skotz: same here so long as it was "support the contest" and not in any way "pay to win"
ddreams: of course
ddreams: just adding some servers during the contest period
UndercoverToad: in the past contests were even sponsored
UndercoverToad: maybe 20 companies. you could select some that may contact you for a job
UndercoverToad: mostly in fr
reCurse: I'm pretty sure you'd need to pay a lot before it starts making a difference
reCurse: Clouds aren't cheap
ddreams: not that expensive either
jacek: you just need to vapor some water. ezpz
darkhorse64: I leave the contest for a long weekend. When I return, the first has a 7 point ELO lead. Is there an anti doping control scheduled for tomorrow ?
Illedan: I would pay for more games through the cg platform for myself
MSmits: they need to monetize us
DomiKo: pay to play?
Illedan: Nah, just for a serious amout of extra games
Astrobytes: darkhorse64 :D
Illedan: Without the new restrictions
Astrobytes: The new restrictions are fking awful
jacek: new restrictions?
MSmits: only for losing cg bench i think. I havent hit the limit just from trying the IDE
Illedan: I hit it the first day
Illedan: when all games were fast
MSmits: hmm, i guess i am slow
struct: I hit during testing
ddreams: I'd pay for everyone to have a better experience.. don't think so many more servers would be needed
darkhorse64: does that counter hiders ?
ddreams: like a 16-core 64gb machine costs like $180 per month
ddreams: err, 180 for 14 days
peerdb: what restrictions?
ddreams: if you play too much in the IDE, you get locked out for a while
peerdb: oh ive never had that happen
peerdb: thanks
Astrobytes: Try tweaking some constants in the IDE and playing matches. Dead in minutes
Illedan: Max 600 a day
ddreams: me neither, but I test mostly locally
peerdb: 600 a day is fine.
assume you spend 16 hours here a day (cus im a nerd), its 38 per hour
peerdb: yeah i dont hit that
Astrobytes: There are other limits. Plays per n seconds, minutes, hours
peerdb: ah okay I didnt know
MSmits: btw.. I rammed out my fastest mcts ever.. took me 2hrs, unfortunately it sucked compared to my beamsearch
BlitzProg: Hey MSmits
ZarthaxX: :(
ddreams: how many rollouts did you get?
BlitzProg: I got My MCTS up too lol
MSmits: 12k or so
BlitzProg: oh my god, first turn?
MSmits: no
ddreams: did you try the seed cheat?
MSmits: 2nd turn
MSmits: whats that ddreams
ZarthaxX: talking with the mcts master here BlitzProg what did you expect
ddreams: don't check if it's legal to seed somewhere
MSmits: mmh for use in mcts?
ddreams: yep
BlitzProg: Well I'm expecting at least to not feel so ridiculously bad in comparison lol
struct: a bug that works
struct: ...
MSmits: ah, i gues that might work yes
JudgeAL: r u guys talking about spring challenge?
BlitzProg: but there I am, with my 330 rollouts turn 2
MSmits: mine was just a SP mcts, not a 2p mcts.
Astrobytes: Yes JudgeAL
BlitzProg: So is mine
MSmits: ah ok
ZarthaxX: MSmits i think that clocks has some good eval, besides using mcts
JudgeAL: ahh i see, how do u train stuff and upload it?
ZarthaxX: i dont think that's the real magic only
MSmits: yeah, i am not saying that mcts wont work
MSmits: i was hoping it could somehow benefit from not needing a midgame eval
ZarthaxX: you mean *will* work
MSmits: like beam does
ZarthaxX: like super easily
ZarthaxX: ahhhh
ZarthaxX: agree
Astrobytes: JudgeAL: Most of the search algos are online, I suspect there is at least one NN though. For that you need to train offline and embed weights
ddreams: how about beam search until midgame?
ddreams: I call it dream search
MSmits: yeah i have been considering combining multiple searches, but dont have the time anymore
BlitzProg: I am mentally exhausted about all of this, but there's some good news: there are 12 hours left
Astrobytes: I think different searches (or evals) for different stages are appropriate for this
Illedan: I have so many timers in my code :D
Astrobytes: lol
MSmits: maybe i should add some
Illedan: if Day > N
if Num3 > K
...
ddreams: lol, the missing piece
ddreams: timers
JudgeAL: ohh, dang just manually copy and paste weights.
ddreams: why do you call that timers though?
MSmits: because he uses 1 eval until time runs out
MSmits: then uses another
ddreams: oh
reCurse: Losses are often suspiciously bunched together.
reCurse: I think servers are up to something
Illedan: Isn't 80 ms enough to solve that?
reCurse: You think I have milliseconds to spare
reCurse: Besides I've seen jumps in the 30ms
Astrobytes: 30 :o
ddreams: it seems like you can afford a couple of losses without losing rank
Illedan: I think he really wants 100% win :P
ddreams: yeah, jk
BlitzProg: MCTS sounds like it won't help me past legend, I'll see if I can set up a BS with the time I have left
reCurse: Sorry wasn't meant as whining
reCurse: I'm just puzzled at what's going on
ddreams: didn't take it like that, was just jabbing at your complete domination
reCurse: When something's too weird I need to know why
MSmits: jabbing is all we got left reCurse :P
Illedan: :D
Astrobytes: hehehe
BoBot: has this kind of domination in a contest occurred before? :D
Astrobytes: Yes.
MSmits: mmh i did halfway through a contest once
MSmits: then the other guy caught up and we dominated together
Illedan: :)
MSmits: and i ended 2 :P
Illedan: Which was that?
MSmits: xmas
ZarthaxX: joli 1 rigth
MSmits: yep
Astrobytes: yeah
MSmits: no nn's though
MSmits: this is something else
MSmits: dont think anyone has done this in an official contest here
Illedan: Wont know for sure until recalc
Illedan: Someone might appear
struct: im still hiding
MSmits: we can see you
reCurse: Good thing there's no PS5 on this one
Illedan: haha
Astrobytes: lol
struct: my current bot is from testing phase
ddreams: ps5?
Zenoscave: lol Do you still fear the dapo with this bot?
reCurse: Just saying
MSmits: reCurse didnt you also have a huge lead on ice and fire?
reCurse: Hmm
reCurse: Maybe
MSmits: I think it was +5 for a long time there too
Zenoscave: Ninja remember Ice and Fire?
ClockSort: i'm getting '-' timeouts at random, are others seeing this
MSmits: maybe not at the very end though
reCurse: Yes
NinjaDoggy: I will always remember ice and fire :(
NinjaDoggy: teccles and I were like 0.5 apart fighting for first
NinjaDoggy: and then recurse comes in with a +4 lead or something :(
Illedan: xD
reCurse: See that couldn't happen now
MSmits: yeah thats what i remember
reCurse: I was batching all day
BlitzProg: Maybe I should write another PHP heuristic
Cappefra: recurse you're insane
DomiKo: he always was
NinjaDoggy: nah recurse is noob, not even 42 points :P
ZarthaxX: famous last words :D
NinjaDoggy: ;)
Illedan: It's like the rest of us is not even trying..
Cappefra: Everybody fighting for 0.2 points and there's a bot leading by +7
reCurse: Here's a better way to put it, you have a better life balance
ddreams: uhh..
Cappefra: when did you start with this bot, out of curiosity?
Illedan: I wish that was the tradeoff
reCurse: Day 1
Cappefra: I didn't see your name until like yesterday
ddreams: he was coding in secret
Numby: reCurse when did you do first submission
reCurse: Yesterday
Numby: LOL
ddreams: struggling to get through wood league
Cappefra: hahahaha
reCurse: I was hiding but did zero batch
Numby: that's so stealthy
reCurse: Just FYI
ddreams: truth, but only because servers were basically broken
JBM: we believe you
NinjaDoggy: zero batch?
ZarthaxX: wow
ddreams: when you say batch, do you mean playing with an alt?
reCurse: Maybe a few hundred of IDE plays total
ZarthaxX: didnt test against anyone??
reCurse: Just to validate it was working
ZarthaxX: beast recurse
JBM: well that's certainly more than *I* lol
Cappefra: so you submitted in wood 2 with your legend bot
reCurse: Doesn't work
Numby: that's so insane :D
reCurse: In bronze though yeah
Cappefra: ok
dloukadakis: you can get to gold by playing randomly lol
Numby: did you submit simple bot to get to gold then submit your main directly?
ddreams: you could have detected wood leagues based on trees
Cappefra: did the bronze boss self-destruct after that?
Cappefra: did it say "woah chill bruh, take it easy jeez"
MSmits: ddreams he needs the codespace
MSmits: no fancy unnecessary code
reCurse: Final bot minus a change or two in bronze
reCurse: I encountered something funny
reCurse: But that's a story for another day
NinjaDoggy: hey recurse you should start working on PM now that first is secured :)
reCurse: I did, inspiration is a bit dry though
kovi: rollover in uint8_t score or sun?
Cappefra: whar's pm again?
ddreams: postmortem
Astrobytes: post mortem (report)
reCurse: Oh if it was just rollover
NinjaDoggy: recap of your bot
Cappefra: aah ok lol didn't get the joke
jacek: trollover?
Astrobytes: That is what I call a solid submit reCurse.
NinjaDoggy: whelp he got to 42...
NinjaDoggy: nah recurse is noob not even 42.69 :P
Vova_s: does anyone know, how the score is calculated?
Astrobytes: lol
ZarthaxX: NinjaDoggy hahahaa
Astrobytes: TrueSkill algo Vova_s
ZarthaxX: astroooo!
Astrobytes: zarthaaaa
ddreams: Illedan resubmits.. let's see
Astrobytes: !
Illedan: Pff
Illedan: Random changes
Illedan: No motivation for 1 week doesn't help :D
Astrobytes: I only went to gold because the Toad insisted
Astrobytes: Was gonna go all in yesterday but, well, yesterday :P
reCurse: I find that hard to believe given your previous statements
reCurse: Did you decide to go all in yesterday today?
Astrobytes: No I got gold the other day
Astrobytes: And I was gonna give up, but was motivated yesterday until the server stuff
1XC: :expressionless:
reCurse: You were motivated? :o
Astrobytes: Just been submitting random crap today
Astrobytes: Not 'motivated' as such. Call it a buoyant enthusiasm.
reCurse: I don't know if it baffles me more
reCurse: :P
Astrobytes: :D
Codechaos: Will the spring challenge be playable after tomorrow? Don't have time to improve my bot any further right now
NinjaDoggy: it will come back as a multi
peerdb: yeah Codechaos, a few hours after it finishes it becomes available again
NinjaDoggy: don't think it's instantly back though?
NinjaDoggy: :O a few hours? :O
Astrobytes: Yeah it will return as a regular multi at some point next day or so
reCurse: Last time it was more or less right after rerun ended
reCurse: If I recall
Astrobytes: Yep.
NinjaDoggy: oh cool
peerdb: yeah after reruns
Codechaos: That's great, thank you!
peerdb: not that I'd know anything about reruns in legend but ye
Illedan: I love how you simply give up when you lose reCurse :O
reCurse: I still do that?
reCurse: FFF
ashiknur: can we see the source code of top participants?
Illedan: Like you assume I wont do anything stupid lol
Astrobytes: No ashiknur
Illedan: https://www.codingame.com/replay/557076400
NinjaDoggy: @illedan nah the bot is just confused because it's never been there ;)
jacek: hm https://img-9gag-fun.9cache.com/photo/aLp47jA_700bwp.webp
MSmits: is that a MTG meme?
Astrobytes: ashiknur: there will be post-contest reports though
jacek: maybe
Illedan: Win 1 game in IDE vs re Curse. 10 loss on submit :rocket:
Illedan: :explosion:
reCurse: I give up
reCurse: Also in real life
reCurse: Dunno how to fix that
reCurse: I tried
jacek: oO
Astrobytes: I think you're safe without fixing it. Multi might be a different story.
MSmits: well if you have some codespace and want my beamsearch ? :P
MSmits: it ought to be better than suicide
1XC: I keep getting fast and I hate fast
reCurse: Maybe with more time but now I'm fed up
Astrobytes: Loads of codespace over here MSmits. Beam me up laddie!
NinjaDoggy: recurse missed a win here? :O
NinjaDoggy: https://www.codingame.com/share-replay/557075088
Astrobytes: :P
MSmits: lol
NinjaDoggy: max seed = 2 ;)
reCurse: Ok this one pisses me off actually lol
reCurse: God damnit
NinjaDoggy: LOL
Queuebee: http://chat.codingame.com/pastebin/afdb0f11-3fed-4e2a-b1c3-8c1886812fa5
Queuebee: sorry wrong chat. I mean. I'm drunk. I mean. Just kidding.
MSmits: count trees like this:
MSmits: int seedCost = __builtin_popcountl(parent.trees[0] & parent.owner);
reCurse: Wait wait wait that game is a complete nonsense
NinjaDoggy: why?
JBM: that would have been my main question actually
reCurse: It said guaranteed win at the end
JBM: do you actually "understand" your AI's games?
Illedan: He wrote an AI to understand it..
reCurse: The answer is I try
MSmits: decreasingly less these days JBM, he doesnt need to :P
JBM: yeah a good position to be in
Queuebee: Oo thanks MSmits
CharmyRosewolf: Okay that's it gold 900+ ftw. I quit!
MSmits: Queuebee thats for a bitboard in c++
NinjaDoggy: reCurse it is guarenteed win... if you plant 1 more seed ;0
peerdb: when doing beamsearch is it better to make the array of gamestates an array of pointers or nah?
CharmyRosewolf: It was fun while it lasted. :D
BoBot: that looked like a pretty extreme sun point collection by Magus
reCurse: One sec panic debug session
linjoehan: Ok time to start for real
Astrobytes: It certainly was
Astrobytes: BoBot
Magus: extreme sun point collection ?
linjoehan: What's the meta before I begin?
NinjaDoggy: ahh yes the typical 7 point lead panic debug session ;)
BoBot: havent seen any other bot play like that yet, maybe there are some exploitative opportunities
ddreams: linje: copy recurse's games
RoboStac: it looks like an mcts that knows it's lost so all moves are bad and then lucks into recurse failing to convert
RoboStac: especially given the debug prints
BoBot: @Magus - your bot was banking 60+ sun points in the shared replay that you drew vs reCurse
kovi: maybe no draw rule
BoBot: I had not seen that strategy before
Queuebee: Oo reCurse I see you joined the competition
Queuebee: When did that happen
Astrobytes: good spot robo
BoBot: @RoboStac or that :D
reCurse: Oh. I see.
Magus: My bot only gather maximum of sun because this is the best way to maximize score + sun/3 at the end
Magus: (since reCurse can't COMPLETE a tree before the end, there is no reason for my bot to complete a tree)
reCurse: There was a bug
reCurse: But the game is also a forced draw
DomiKo: :scream:
reCurse: Planting a seed costs 3 energy = 1 less point
reCurse: So... all good I guess?
kovi: 2
kovi: 11-2=9
DomiKo: yeah
reCurse: And I'm a monumental shame
JBM: why are you almost always red
ddreams: red is closest to recurse than blue, doh
Astrobytes: It's the allergies
Astrobytes: ;)
reCurse: I'm blue on that replay
JBM: i'm on your curent agent's last battles page
ddreams: jbm, it's completely random
ddreams: on the left means he's always winning
JBM: oh right
JBM: i forgot how to read that one
JBM: thanks ddreams, i needed the readjustment
Illedan: reCurse having 10+ people debugging his games
Astrobytes: How many years JBM... :D
JBM: too much, and counting
Astrobytes: hehehe
reCurse: I'll only fix that bug if there is another game where it happens
NinjaDoggy: doesn't planting a seed cost 2 there?
Astrobytes: Illedan - you're missing the public debugging/rubber ducking session. Fix it.
reCurse: Yes yes
reCurse: I can't math
NinjaDoggy: because it would be your 3rd seed?
JBM: ask pb4
JBM: he's available this time
reCurse: Pruning got a little too aggressive
reCurse: It literally can't seed
ddreams: that's my problem too, keep pruning the winning moves...
peerdb: do you guys prune after each turn or after each day?
JBM: every chance i get
peerdb: how far do you prune?
reCurse: In his case he may have pruned the contest
peerdb: im pruning my sanity tbf
Astrobytes: I always prune to just above the new bud.
MSmits: well you're a professional
Astrobytes: lol
reCurse: So it turns out my random thought of the other day kinda matters
MSmits: what was that?
reCurse: Assuming you have perfect play (not making implications here, bear with me for this example) and the opponent doesn't
Astrobytes: Which one? You had a few iirc
reCurse: You start in a position solved as a loss
reCurse: How do you play?
MSmits: oh right, i've had some time to think about that as well
MSmits: did not find a solution myself
MSmits: well some imperfect ones
reCurse: I have a weaker variant of that problem I think
Astrobytes: So maximising/drawing out the game length (As in Oware) doesn't work then
MSmits: your bot hasn't learned to discriminate between bad states and really bad states
RoboStac: I've always gone for the move with the most visits as usually that means the loss was hard to solve and maybe they'll miss it. Doesn't really work here as it just ends up spamming seeds
reCurse: But that doesn't mean it's the good way
reCurse: The theroetical best way is what the opponent expects least
RoboStac: yeah
reCurse: But defining that is...
MSmits: what you can do is play the moves that have the most losing replies
Astrobytes: A problem requiring more than 11 days
RoboStac: I guess you'd possibly need to analyze it more like chess where keeping it as complicated as possible (keeping pieces on the board to have more options) is probably better?
Astrobytes: MSmits that was what I was getting at with the Oware comment
RoboStac: though again hard to apply that here
MSmits: yeah
reCurse: Indeed
Vova_s: you can go for the path that will grant you less difference in points
reCurse: Because some of the maps are god awful
MSmits: Vova_s he might not be measuring points
kurnevsky: There is a dynamic komi technique, but not sure how well it will work here :)
kurnevsky: Allows to maximize score in loosing positions.
RoboStac: yeah, but that probably doesn't maximise winning chances
reCurse: Supervised learning on top 10 maybe lol
reCurse: The contempt NN
MSmits: mmh thats sort of what i do :P
Astrobytes: It might work if the opponent blunders, even to force a draw
MSmits: your best bet is to have a bot so good, it wont get into losing situations
reCurse: See my previous point about god awful maps
ThunderbirdOne: ah, just my luck
ThunderbirdOne: forgot about this comp
Astrobytes: yeah
RoboStac: yeah, but that only works for one of us
ThunderbirdOne: i remember 11hours before it ends
MSmits: right, only one person can have a bot that never loses :)
Astrobytes: Wondered where you went ThunderbirdOne
ThunderbirdOne: haha
reCurse: A swindling AI would be pretty funny
ThunderbirdOne: i got distracted
ThunderbirdOne: and then discord became a thiing
ThunderbirdOne: and life
Chainman: my heuristics doesn't really work with beam search
ThunderbirdOne: haven't really been coding much outside workhours either
Astrobytes: You were fixing your house last time iirc
ThunderbirdOne: aye, ground floor is pretty much done
ThunderbirdOne: saving for all the rest now lulz
ThunderbirdOne: shit is so expensive :')
AntiSquid: hi ThunderbirdOne how's life ?
ThunderbirdOne: pretty good actually
ThunderbirdOne: anything i missed here? :D
AntiSquid: awesome, glad to hear
Astrobytes: haha, nice, you'll get it done eventually
AntiSquid: you missed a lot of frustration imo with this contest, pretty lucky actually to come when it ends :P
ThunderbirdOne: oh lol
ThunderbirdOne: how come?
AntiSquid: it's my opinion .
ThunderbirdOne: lulz
AntiSquid: it's not a fun game imo
ddreams: my frustration was only about the servers
Astrobytes: imo it's the contest that no-one really likes but did it anyway.
For me.
ddreams: game was nice enough
ThunderbirdOne: Haven't seen our Russian friend in the chat tonight... (I even forgot his name)
Astrobytes: Personally speaking.
AntiSquid: hah that's me astro .
ddreams: I liked that you could get far without search
ThunderbirdOne: Wood battles are so slow btw
ThunderbirdOne: holy smokes
wlesavo: Astrobytes thats basically summarize my view on this :slight_smile:
AntiSquid: madao is MIA most of the time if that's who you mean
ThunderbirdOne: right, Maddy
ThunderbirdOne: MIA, that's unlike him
ThunderbirdOne: good to see the regulars are still around though
ZarthaxX: hey ThunderbirdOne!!!
AntiSquid: he's usually here before contest starts and then disappears
ZarthaxX: oh wlesavo alive :O
reCurse: We're up to a count of 2
ThunderbirdOne: i havent been here in a year or something
reCurse: Stop it
Astrobytes: wlesavo: I'm glad I have a few supporters of this view lol
ThunderbirdOne: Hiya ZarthaxX!
ZarthaxX: how are you maaaaan
ThunderbirdOne: pretty good, found some spare time and noticed the comp :D might as well
ZarthaxX: doing it?
ThunderbirdOne: i'm just glad people still remember me tbh :')
ZarthaxX: last second legend?
ZarthaxX: cant forget the rocket
ZarthaxX: :)
ThunderbirdOne: going through the first league now :')
ZarthaxX: haha ok
ThunderbirdOne: winning every battle, but its slow so... just waiting
ZarthaxX: rush gold :)
Astrobytes: Count of 2 reCurse?
wlesavo: ZarthaxX i missed first half, and im pretty slow, so yeah, half alive i would say =)
reCurse: Say the name 3 times...
ThunderbirdOne: going to bed in a bit i think... conference next 2 days (digital though)
Astrobytes: reCurse: Ah gotcha :D
Zenoscave: we really should stop pinging the curse
ZarthaxX: wlesavo oh rushing legend?
reCurse: I wasn't worried about that
AntiSquid: reCurse so it's a 3rd time
ZarthaxX: ThunderbirdOne sounds exhausting :P
Astrobytes: No you missed the point
Zenoscave: * <--- point
o <--- my head
ThunderbirdOne: what league is CatBoss btw?
Astrobytes: I'm gold, thanks for asking
reCurse: Most awesome boss name
wlesavo: nah, i gave up on legend, cant improove my beam search anymore, may be ill finish it after reading some pms. top gold is fine
reCurse: Bronze
ThunderbirdOne: oh okay
ThunderbirdOne: hard to get past bronze?
ThunderbirdOne: doable in a couple of hours?
Astrobytes: No and yes
ThunderbirdOne: alright, nvm then :')
Chainman: crap my beam search is way worse than not doing it rn.
Astrobytes: You can definitely get Silver at least
wlesavo: i got gold with completely broken shadows
Astrobytes: I got gold without any shadows
wlesavo: broken worse then none :smiley:
Chainman: what are shadows?
Chainman: :stuck_out_tongue:
wlesavo: anyway, good night, still was quite fun to be around this week
AntiSquid: ThunderbirdOne just throw in a bunch of random conditions, make sure 50% of them are bugged and you will promote easily
ThunderbirdOne: cool cool
ThunderbirdOne: im in Wood 1 now...
Chainman: I did the beam search, but according to my eval, it doesn't know what to do. Like I need to redo my eval completely.
ddreams: Chainman make sure your beamsearch isn't bugged first
Chainman: it's not it's choosing correctly
Astrobytes: gn wlesavo
Chainman: but i had a bunch of if, else in my eval that would depend on if it was the beginning, middle or end of game.
ddreams: when choosing the action, you should only compare full days, not same depths
ddreams: ie. choose best action from wait nodes at same day
ddreams: comparing nodes for different days is not working well
Chainman: oh crap, that's genius I believe, thanks
Chainman: yeah because your heuristics can depend on the day.
Chainman: especially mine.
ddreams: the IDE should have a swap sides button
Ashargin: I've been stuck in top gold for days. Can't find the right heuristics to beat the boss...
ymoukhli: same
ddreams: just fixing bugs and misunderstandings was more than enough to get me past the gold boss
ddreams: was struggling next to it for a long time too
MomoLand: hey
MomoLand: any interesting projects out there you doing ?
Chainman: using beam search to beat gold boss?
Numby: So I have consistently this 55ms spike on turn 6 or 7 and I cant figure out what causes it :( I tried timing each of my big functions and sometimes it's one function sometimes its another where the spike occurs. It's also weird that this only happens early in the game
MomoLand: what gold boss guys ?
PiterYeh: ddreams i with i could send your advice about day nodes to my past self 10 hours in the past. I'll try that now, thanks
Numby: Because of the spike I gotta keep my timeout to 50ms or I go over the limit
ddreams: gl PiterYeh, still time :)
Chainman: the day nodes is key, in fact. I think changing depth to number of days.
ddreams: you can rest when you're dead
ashiknur: what's this?
An error occurred (#407): "You reached the limit of plays for a period of time."
Ashargin: It's pretty self-explanatory
ddreams: means you need to test locally for a bit
Zenoscave: do you order moves in a certain some order?
Chainman: Those who are near beating gold boss do you win sometimes?
Ashargin: Yeah I win like 30% of the time maybe
Ashargin: I'm not sure what I should improve :/
AeroFufel: Is it possible to make the chat bigger, in a separate page maybe? a hint anybody?
Pduhard-: i was beating gold boss 75% times but was loosing against other top bots, stuck near top 10 for a while before the upgrade which got me legend
ThunderbirdOne: lets see if this gets out of wood 1 :')
Chainman: yeah
Chainman: my eval was giving score according to the day, and the actions, should it be giving score according to the gamestate with beam search?
Chainman: hmmm I think it should work with actions.
Chainman: just want it to look 2 days in advance maybe.
linjoehan: ThunderbirdOne I did movelist[max(movelist.size()-1,1)] to get out of wood
ThunderbirdOne: oh im just upgrading the biggest tree i have enough sun for at this point lol
linjoehan: or was that min no it was min
Illedan: ddreams :)
ddreams: are you coming for me?
Illedan: Above atm, but ranks are very close around us
ddreams: we had one match so far, 152-115
ddreams: I am almost ready to resubmit too
Illedan: :scream:
Illedan: I just yoloed some params
ddreams: I changed nothing, but drastically sped up the algo
ddreams: not sure if that translates into playing strength
Illedan: Depends how you use that speed
ddreams: I'm beam searching for 90ms, so it could help look a bit further
Illedan: True, unless you prune the good moves early :)
ddreams: yeah, I could perhaps widen the beam
Michael_Howard: What's the best tactic for seed wars like this? I mean, it's annoying both sides equally, not sure if I should do anything in particular. https://www.codingame.com/replay/557125733
ddreams: well, in the end you won
Michael_Howard: Yeah but it sapped my arena bot's confidence & now he's run out of the Gold top 100 & won't go back.
AndreMarasca: symmetrical game, I thought I was going to get stuck throwing seeds in the middle
Chainman: I don't think you can use beam search if your eval function is scoring actions based only on gamestate day and not the gamestate.
ddreams: why not?
ddreams: day is part of gamestate imo
ashiknur: I was at 650 in gold and I submitted an upgraded code for a better hope and I tucked at 1235. :persevere:
Chainman: yeah but that is the only thing I'm using from gamestate.
Chainman: it's like my algo sees a really good action on day 3, but there are multiple paths there and it chooses the wrong one.
Chainman: let's say I'm looking 2 days in advance.
Michael_Howard: AndreMarasca yeah kinda kaleidoscopic.
Chainman: normally on day 1 I'd prefer to grow tree to size 2.
Chainman: But it sees that on day 3 there is a equivalent good score for the path that starts with just waiting on day 1 :(
ddreams: why do those have the same score if you think it's worse?
ClockSort: i had that problem in fall2020 challenge. it would see a good state in future and start with a REST action.
ClockSort: my simple solution here was to just break ties in GROW -> COMPLETE -> REST order
Illedan: Why not add value to the beams along the way?
Zenoscave: what do you mean ille?
ddreams: as in longer sequence is better?
Illedan: As in doing something earlier is more score
Zenoscave: got it.
PatrickMcGinnisII: what's recurse's secret?
ddreams: win all the fights
ddreams: sshh
BoBot: it is obvious by the name, isn't it? ^^
AeroFufel: A friendship with the tester and an early start )) or kidding ))
ProG_r_aM_m_er: I think he found the finite number of possible actions ;)
ddreams: I find that incredibly hard to balance, Illedan
ddreams: often waiting really is better
ClockSort: you have 1000ms during the first turn, so just explore all possible future game states during it
Astrobytes: *just*
ClockSort: is my bot getting worse or is top-20 getting a lot tougher over the past 6 hours?
Chainman: waiting a lot
natofp: what is happening if you submit code 5 mins before end of the contest and not all battles are plated?
natofp: played*
RoboStac: it'll finish playing all the battles
ClockSort: they get plated after deadline
ddreams: you risk it all
natofp: ok
natofp: just battles get played right?
natofp: and after this
natofp: this will be the final position?
ddreams: I did that in legend once, top 20 or delete yolo, and it just kept timing out and I ended up at the very bottom
BoBot: you might win the contest with 100% winrate ^^
natofp: xD
RoboStac: yes (there are some extra battles played between the top 100 in legend once all bots have finished)
natofp: so there is a cutoff
natofp: like
natofp: > 100 players are fixed
natofp: and <= 100 players are playing to permute their order possibly?
RoboStac: I'm not actually sure what the rules are on the 100 boundary
ProG_r_aM_m_er: when i beat silver, i already beat half of gold
RoboStac: my guess is if you are just outside the top 100 your score is still affected as you can be used as an opponent
ClockSort: @PRO_G_r_AM_m_ER when i beat gold boss, i also beat 70% of legend too :P
ProG_r_aM_m_er: if you play against yourself and you sometimes win/loose, in the beginning shadows determine victory
ProG_r_aM_m_er: yeah, well maybe in the future I will have more time and better strategies to work out
Chainman: interesting, it thinks I can grow infinite times on my second turn.
ClockSort: maybe it is right
Illedan: Cached the cost?
ddreams: forgot to set it to dormant
ddreams: early on I found I planted several seeds at the same location.. forgot to mark it as taken in the anytrees-bitboard
Chainman: yeah I forgot the dormant
Chainman: and my generator is generating impossible moves
Illedan: The feeling when you hit play limit while tweaking some param on the last night -.-
Illedan: (╯°□°）╯︵ ┻━┻
Chainman: couple o' bugs up in here.
Chainman: ooooh no
ddreams: that's crap illedan
Astrobytes: Param tweaking in IDE pissed me off sooooooooooooooo much this contest
Illedan: Yeah
ddreams: is it allowed to share bot executables?
Zenoscave: I don't think so ddreams
Astrobytes: And it was introduced before it too. I thought I was going nuts when I was tweaking my other multi bots
ddreams: don't see how that is cheating though
Astrobytes: Between users I suppose ddreams but not publicly
Astrobytes: And not during contest I'd say
Zenoscave: and definitely not the source. just the binary
Zenoscave: Although that doesn't stop decompilation i guess
Astrobytes: ^
ddreams: yeah, would be a stripped binary
Astrobytes: Yeah, after contest that would be fine imo
ProG_r_aM_m_er: https://www.codingame.com/replay/557155732 vs https://www.codingame.com/replay/557156282 random growing for gold bot :D
ddreams: after contest you could share code I guess
Astrobytes: No
Zenoscave: No, Never code.
Astrobytes: Snippets yes
Zenoscave: This isn't commonly allowed in competitive code. Snippets are ok, like astro says
Zenoscave: as long as it isn't too much
ddreams: but streaming is encouraged tho?
Zenoscave: For lower leagues.
Astrobytes: Unless it's legend
Zenoscave: You'll note many streams stop at gold at the very most
Astrobytes: OR high gold etc
Chainman: I want to see recurse stream :innocent:
Astrobytes: He has one on youtube
Zenoscave: after locam i doubt that'll happen lol
Astrobytes: There was definitely not another stream about another game that isn't available.
Chainman: I need to watch some videos and stream after contest actually to learn.
ddreams: I want to see him stream while coding his bronze bot
Illedan: I learnt more from streaming than watching streams :P
Chainman: what is locam?
ddreams: Can you elaborate, Illedan?
Numby: viewers give you tips?
Chainman: https://www.youtube.com/watch?v=BU9b445CpaM
Illedan: I watched my own streams to learn more about myself
Chainman: he has two videos
ddreams: that's an answer I didn't expect
ddreams: what did you learn?
Illedan: About my problem solving skills while under pressure of the stream. And how I tackle it.
And about my english speaking
Zenoscave: I remember your english was great from my pov.
Zenoscave: despite mic problems ;)
Illedan: shhh
Astrobytes: Tbh Illedan, your English speaking has been rather excellent. Volume skills however....
Zenoscave: now help my stupid eval
Illedan: Yeah, don't get me started on this new Mic i bought which is no good
Astrobytes: -3suns
Zenoscave: I swear the moment I get a good eval. top 20
Zenoscave: it'll be on next sunday
Illedan: I wish
Illedan: I guess I'll accept 2nd NO :(
theycallmedavid: End of game 100 50.00% 50.00%
https://i.kym-cdn.com/entries/icons/mobile/000/023/397/C-658VsXoAo3ovC.jpg
Astrobytes: Weird game. I still don't like it.
ddreams: just wait until I submit my new bot and destroy myself
Illedan: ^^
Astrobytes: lol
Zenoscave: The same bot I submit varies from 100-200
Zenoscave: there are only 200 in Legend btw
ddreams: ok, I'm submitting.. how long does it take for a run now?
Zenoscave: about an hour
Zenoscave: My guess
Astrobytes: There is always that effect though zeno. Gold and legend, every contest.
Zenoscave: That wide of variance though?
ClockSort: a quarter of a run is good enough to get placement if you aren't near the top
ddreams: here goes then
Astrobytes: With many active submits possibly
ddreams: only 340 now
ddreams: so going down
Illedan: Night time is upon us
Illedan: And I was supposed to go to bed 40 min ago
ddreams: 5 out of 10 initial battles
Chainman: I hope to get top 300 sooon
ddreams: no sleep until you beat euler, remember
Zenoscave: 3/ 10 :/
Illedan: I gave that up last sunday :P
Illedan: But I can't accept 2nd NO. So here I am
Astrobytes: The duties of life Illedan
ddreams: hehe, you're 1st now
Illedan: But this blocked IDE games will kill me
ddreams: enjoy it
Illedan: Screenshotted, printed and put on my wall
CouscousAgha: what could happen if the timer hits 0 and someone is still submitting
ddreams: we're almost as many competitors as from italy
Zenoscave: CouscousAgha you're immediately hated
CouscousAgha: how so
Chainman: lol
Zenoscave: It won't let the legend rerun until all submits complete
Astrobytes: ^
Zenoscave: And pray no one knows your address if you submit last second in gold near top
Astrobytes: hehehe
Zenoscave: Because that could mean a promotion and re submit in legend
CouscousAgha: all legends will automatically resubmit?
Zenoscave: Of course I aim to do this every time.
top 100 legends get a re-ranking
emagnidoc: hello guys, how can i approve a contribution from another user?
Astrobytes: Yes. After the submits finish, the top 100 recalc
Zenoscave: be a higher level emagnidoc
Whatar: lvl 20 required
Zenoscave: to keep from spam
RoboStac: you can approve clashes after 50 public clashes can't you?
emagnidoc: okay, thanks ... i was just wondering because i have the quest ... but couldnt find a way to solve it
Astrobytes: Something like that yeah RoboStac
Zenoscave: I think so
Whatar: but you also need to be lvl 20 I think
RoboStac: no, lvl 20 for puzzles
RoboStac: 50 public clashes for clash
Whatar: oh, ok, so I should be able to do it
Whatar: let me check
emagnidoc: @Whatar ... its the blue branch on the bottom right
Zenoscave: emagnidoc just don't approve crap to complete a quest.
Zenoscave: Put effort into finding a good one
Astrobytes: Looks like Smits gave up, unless he has something up his proverbial sleeve.
What about you RoboStac?
Whatar: I don't have the button to approve it
emagnidoc: @Whatar .. .same here thats why i was asking
ddreams: looks like my new and improved bot is getting beat
Whatar: oh, probably because its wip
Whatar: let me search better
RoboStac: Whatar - are you sure its not wip?
RoboStac: ah
RoboStac: https://www.codingame.com/contribute/view/6663b740cc323c93fdacb7cdb7c58ffb8402 ?
Astrobytes: Standard procedure for CG bots ddreams :)
RoboStac: I'm just trying random things to try and get back into the top 20 Astrobytes
RoboStac: though its so slow I've no idea what is really happening
cegprakash: someone plz share me a replay between me and Robostac
cegprakash: someone in legend
RoboStac: you aren't ranked high enough for me to find you
Astrobytes: Fair. It's been ridiculously slow this time around.
Astrobytes: It's not just me is it?
RoboStac: I gave up on my rewritten version and went back to wednesdays version as I couldn't make duct work
linjoehan: I've never been able to get duct to work
Whatar: any Idea on what my bot is doing wrong?
Whatar: https://www.codingame.com/replay/557169150
linjoehan: well not fast enough to work
Chainman: I think I'm gonig to call it quites on beam search
Chainman: https://www.codingame.com/share-replay/557174647 wth
Astrobytes: Whatar: Firstly - it's losing. That's the wrong thing to do :P
ddreams: I agree with Astro here
ddreams: Stop doing that
Whatar: True Astro
Whatar: but that, I don't know how to fix
Illedan: I'm off. GG everyone. Cya next contest ^^ :)
Astrobytes: Honestly, count me out of serious discussions right now. My brain hurts :D
Chainman: bye
ddreams: gn illedan
Whatar: gg
Chainman: I think Whatar you were close though
sasquatch: Question for someone in legendary. How do I get to play MiyazaBoss in the gold league? My best bot ranks about 50th in gold
Astrobytes: see ya next week Illedan :)
Chainman: have good dreams illedan
Chainman: dream of your bot winning
ddreams: get to top 5 or so sasquatch
RoboStac: Whatar - you gather less sun on 17 of 24 days
sasquatch: I see! Thanks ddreams
RoboStac: mostly only by 1/2 (though day 12 is 9 vs 17) but it adds up
ddreams: oh right, I removed all my move pruning as an experiment
Whatar: yeah, my fat totoro tells me the difference in sun gains and I am a lot negative by the end of the match
ddreams: is there any time it's better not to complete first thing in a round?
ddreams: maybe if there's a seed location competition
RoboStac: if you wait till opponent is asleep you might be able to get multiple nutrient bonuses in a row
ddreams: you think the opponent decides to complete if you do it first thing in a round?
ddreams: clocksort resubmitting too, I see
ddreams: 6/10 initial battles this time
ClockSort: i think i'm making my bot worse with every iteration :sweat_smile:
ddreams: I wonder why the game doesn't start you with 0 trees...
ddreams: instead we have this asymmetric hell
reCurse: How would that solve symmetry
ClockSort: yeah @ddreams it's an unfair start
reCurse: You'd end up with a PR1 situation
reCurse: Nobody wants PR1
ClockSort: but you have to start somerwhere
ddreams: pr1?
ClockSort: what's PR1?
reCurse: Pacific Rift 1
elderlybeginner: should be at lease rabbit round
elderlybeginner: least
ClockSort: not familiar with that situation
Magus: Platinum Rift 1 ?
reCurse: lol
reCurse: Yes sorry
reCurse: Mind went to strange place
ddreams: it's impossible to get started?
ddreams: if both players want the same starting spots
reCurse: Everyone places on the best spot so you either have to suffer instant defeat or compromise to a worse position
ddreams: right
reCurse: There is no solving for symmetry with this game design
reCurse: You need a new game
Magus: PR1 is a 1v1v1v1 game and suffers a lot from "shifumi" games. Like if another player wants to spawn at the same place as you; you just lose the game and you can't do anything about it.
ClockSort: yeah the sun is screwy, literally
ddreams: yeah, that sucks more than this
reCurse: PR1 had the same issue in 1v1
PatrickMcGinnisII: recurse skips a seed to avoid a bad position, and next turn seeds next to a tree he's gonna complete...i';m like wtf
ddreams: all battles run twice on the same map then?
reCurse: Yes that would be the alternative
reCurse: However it costs more
reCurse: They used to do that
reCurse: But that makes everything twice as slow
reCurse: You want 2x slower submit?
ClockSort: sign me up
ddreams: Yes, and more servers
Astrobytes: You were simply being very Pacific reCurse
reCurse: I think my mind was on Pacific Rim
reCurse: Not sure why
Astrobytes: woo chat scroll, sorry
Astrobytes: Yes, looks that way
elderlybeginner: why it's makes submits x2 slower? Amount of fights could be the same
reCurse: Maybe some inner need to have giant robots smash giant aliens
Astrobytes: Never a bad thing imo
reCurse: No because the result of a pair of games still counts as one game
ddreams: why?
reCurse: You need 2 games to determine the final result
RoboStac: either way it still only gives you 45 opponents instead of 90 so the rankings will be much less stable
reCurse: Well I guess that's the crappier alternative
ddreams: I think adding a few servers seem like a good solution
PatrickMcGinnisII: I'm unstable, I'm 1/2 the square of the root of the 2nd participle away from death by analysis
Astrobytes: I think you should tell CG this ddreams.
ddreams: Like joining this contest has cost me maybe $10-15k
Astrobytes: ?
Whatar: healthcare?
ddreams: not working for money
PatrickMcGinnisII: google ads, rofl..i have no idea
ddreams: my clients pay me a lot
PatrickMcGinnisII: norwegians have OnlyFans?
Astrobytes: Welllll.... I see your point. But... fun needs no repayment
linjoehan: lol joining this contest has cost me sleep
Westicles: ddreams, you dance on the webcam too?
ddreams: yeah, but I'm totally willing to pay for a bit of server time
ddreams: Westicles when I forget they are watching
Astrobytes: I'm with you on that yeah
Chainman: :sweat_smile:
linjoehan: Ok first version of sim complete now where is that MCTS... just copy and paste and it should all work...
Chainman: nice, I think I'm giving up on beam search, it's too hard for me.
Chainman: I need to read/learn more
PatrickMcGinnisII: My daughter is suggesting we play Minecraft, why does that sound less insane than poking wood sprites with hot pokers
Chainman: hmmmmmm
Astrobytes: Anyways, I'm outta here. Well done and good luck wherever it applies
Astrobytes: RoboStac: submit looks good
PatrickMcGinnisII: glhf Astrobytes
ddreams: ciao astro
Chainman: have a good on
Chainman: one*
RoboStac: Chainman - hypersonic is a good beam search game that is quite a bit simpler (no turn/day complexities) if you want to try it somewhere easier
Westicles: Hah, look at tbali going for the language badge in objC
DomiKo: or Fall Challange
Chainman: okay thanks,
Chainman: I think I need a simpler game yeah for starting.
Westicles: And pardouin for clojure
Westicles: Looks like Bash is the easiest one to go for now
ddreams: what about pascal?
Westicles: Some guy has gold already
cegprakash: wth
cegprakash: I started to win boss locally
cegprakash: legend?
ddreams: gl
cegprakash: added 1 flag in constructor
cegprakash: and that's changing my win rate drastically
cegprakash: wth
linjoehan: Man my MCTS only completes 25 games in time. Plan B then.
ClockSort: get really lucky.
solistice: ugggggh 7th in silver
solistice: would really love to hit gold
ddreams: share a replay, sol
solistice: https://www.codingame.com/share-replay/557205507
ddreams: what immediately stands out is that you're shadowing yourself a lot
Whatar: yeah, even just planting your trees one block away can bring improvements
Whatar: one more I mean
Whatar: but if you can avoid shadowing yourself entierly, that would be good
elderlybeginner: solistice the main rule is not to seed in the line
ddreams: yeah, but just looking the neighbors of a seed if you have any trees there would probably improve it a lot
ddreams: like whatar said
ddreams: it's not so hard to go further, but...
ClockSort: my MCTS has a problem when the opponent has 50 suns. it's not lucky enough to try the sequence where none are spent, so it thinks it is winning even when it will lose by a hoarding-sun opponent. any suggestions?
Bozo: for those who implemented a MC tree search, what is its max depth ? just out of curiosity
ClockSort: @Bozo i play to endgame
Bozo: nice
ddreams: ClockSort give a boost to wait actions?
Bozo: my tree search is not a MC, and is quite short depth
Bozo: its a shame, i didn't know MC before tonight
Bozo: ;-)
ddreams: or ensure runs where no complete actions are taken until last day perhaps
JohnCM: clocksort you can artificially increase the explore portion of the no action component in end game turns (day 19-24)
ClockSort: :bulb:
JohnCM: or if the opponent has no lvl 3 trees remaining, then give a huge boost to the explore component of the wait action
JohnCM: cause typically ppl cash out all their trees they're not going to build new ones already
ddreams: hoarders typically have many level 3 trees though
JohnCM: haha then this strat doesn't really work
are there a lot of hoarders in legend?
JohnCM: cause in gold it is typically the fast cash out strategy
ClockSort: when i am the opponent then everybody is a hoarder
ClockSort: i cash out so fast that you must hoard to win :D
ddreams: yeah, you probably induce it by grabbing all the nutrients
ddreams: I think I finally found an actual eval improvement
cegprakash: timer is red
cegprakash: I'm doomed
ddreams: still 8 hours, come on
ddreams: lol Westicles
ddreams: going for pascal
PatrickMcGinnisII: finally beat the darn gold boss
ClockSort: congrats @PatrickMcGinnisII !!
ddreams: grats :)
ddreams: just in time
PatrickMcGinnisII: only once gheesh
PatrickMcGinnisII: be back later, groceries
PatrickMcGinnisII: tx tho
Westicles: ddreams, I'll have to settle for top3, I don't have a silver
ddreams: I wonder why I have a top 3 in a language ranking
solistice: will cobol ever be supported?
solistice: just asking for a friend
sgalasso: How the hell is it possible that some bot manage to have a good base score AND cumulate sun at endgame, just how
ddreams: lots of sun I guess
sgalasso: Yes i know, but I mean my algo try to use optimaly sun I ahve during the match, theirs do it plus they accumuate x) I have no clue about how they do it
ddreams: compare how much sun you and they have accumulated on the first complete
sgalasso: almost same, it's mainly at the endgame
ddreams: when does it start to differ?
sgalasso: ~17 in average I would say
ddreams: perhaps you're undervaluing the bonus values for score
solistice: sweet, I'm gonna reach gold! :D thanks for the neighbor seed tip!
ddreams: hard to guess
ddreams: np
peerdb: well, here goes nothing! lets push
.TN: ssa
.TN: hello
AlexanderGuk: where top players usually post their PMs? would like to read some
ddreams: forum
AlexanderGuk: just as replies to some admin thread after the contest?
ddreams: https://www.codingame.com/forum/t/spring-challenge-2020-feedback-strategy/184113/62 for example
AlexanderGuk: cool, thank you!
theycallmedavid: I'm getting better at beating myself. I wonder if it will work on others too.
aCat: yes
aCat: they will be better at beating you too
peerdb: :sad:
theycallmedavid: Yes, against the boss that is exactly what happens
struct: hello
struct: im back for the all nighter
struct: :'(
ddreams: then there's two of us
ddreams: I wish my improvements actually improved things
struct: ping me if any action is needed in chat
struct: it will be minimized
ddreams: you got a plan?
struct: a plan?
ddreams: for the last 7 hours
struct: nope
struct: :D
ddreams: spamming tweaks until the IDE locks you out?
struct: im not using ide to test stuff
struct: I run it localy
ddreams: I gave up trying mcts when I thought I saw some eval improvements
ddreams: but they don't pan out
Lucky30_: i am tired, and tomorrow i have an interview, should i keep going with the contest, or just give up
Cjx_1: this contest prly wont pay ya friend :P
ddreams: clearly the best option to bomb the interview because you were so passionate about coding
Cjx_1: lol
peerdb: submit you contest code for the interview
Lucky30_: hhhhhh
Lucky30_: you guys sure now how to cheer up someone
Lucky30_: i guess i could try my luck at legend after the contest
ddreams: if you were in top gold, I'd probably have stayed if I were you :)
Lucky30_: 764 gold =>top 1 does not sound like possible right now
cegprakash: I am unable to test in local arena code against boss it automatically switches to IDE code
ddreams: but you're probably quite far off
cegprakash: is it a feature or a bug
ddreams: maybe you're pushing replay in same conditions
cegprakash: yes
ddreams: same conditions include the players
peerdb: ^^
cegprakash: I want to test under same conditions with arena bot
cegprakash: :(
peerdb: you can save the seed
ddreams: then go to options and choose manual
Lucky30_: anyways, i tryed hard in this contest, but not hard enought, next time i will do better
Lucky30_: goodluck everyone
ddreams: gl to you in your interview too
Lucky30_: thanks ddreams
stefan-gg: im so stuck in silver do you guys have any general tips ?
ddreams: don't seed next to your own trees
peerdb: gl everyone, might check in in the morning, but i need some sleep
stefan-gg: is it worth to occupy the green area early on ?
ddreams: no, it gets shaded a lot
peerdb: it can be stefan-gg, but most importantly, space your trees evenly throught the board
cegprakash: I'm at 40 points in gold and boss is at 47.. is there anything amazing I can do to get 7 points :P
peerdb: pay thibaud
ddreams: win 100% of your fights
ddreams: that's be amazing
ddreams: and give you 7 points or more
stefan-gg: yea you are right but this is my first competiton so i was kinda scared a little bit
peerdb: cegprakash is it just me or are top 150 gold mostly 50-50
cegprakash: ddreams funny
cegprakash: I'll laugh after contest
ddreams: :)
AllYourTrees: omg i copied recurses strategy and have won first 18 games of submit in gold :O
ddreams: what's his strategy again?
AllYourTrees: my previous best was 500 in gold lol
PRO_G_r_AM_m_ER: what is his strategy :D ?
Pick8chu: yeah what's his?
Numby: teach us to copy
AllYourTrees: watch his games :)
AllYourTrees: aww i lost some, maybe not a perfect copy haha
Pick8chu: he barely put seeds on the middle
ProG_r_aM_m_er: I think I cannot improve my code anymore top 400 good enough for my if then else code
Numby: i could use some good if-else for my pruning right now
AllYourTrees: ya i gave up on my approach
ddreams: it's not a dumb idea
CouscousAgha: my if then code made it to top 115 legend
struct: nice
CouscousAgha: im satisfied with that
AllYourTrees: nice!!
Pick8chu: you made your if then to top 115?
Pick8chu: that's really impressive
CouscousAgha: yes
Numby: teach us heuristics :D
CouscousAgha: and this is my first coding game challenge ever
CouscousAgha: Numby notebook and a pen
CouscousAgha: and start brainstorming
CouscousAgha: you will notice a pattern and just prototype it
CouscousAgha: notice repeated processes
CouscousAgha: turn them into functions
CouscousAgha: test and see what u did wrong
CouscousAgha: and repeat
Cjx_1: ok, here towards the end. Who cares about highest elo. Who has the most dubmits. Ik its not me with a measley 80
Cjx_1: *submits
ddreams: 26 submits
KelvinAndHubbles: ... 149
struct: 27
AllYourTrees: wow this might actually do it
codingWhale: Congrats. :)
AllYourTrees: OMG last battle was loss against boss but IT DID IT
Cjx_1: <3 big grats friend
ddreams: gz
struct: grats
Queuebee: allyourtrees
Queuebee: congratz!
AllYourTrees: tyty!!
CouscousAgha: congrats!
Queuebee: nice name for this challange :P
CouscousAgha: welcome to legend
AllYourTrees: haha i may or may not have changed my name just for it :D
AllYourTrees: ty CouscousAgha
cegprakash: AllYourTrees what did u add
AllYourTrees: i copied reCurse's strategy as best I could >.>
cegprakash: oh come on :P
cegprakash: now plz share a replay
cegprakash: with me vs recurse
ddreams: lol Illedan sneaky resubmits
AllYourTrees: https://www.codingame.com/replay/557297550
ddreams: not bad
BrunoFelthes: Are you using search, or most of heuristics AllYourTrees?
AllYourTrees: only heuristics
struct: bruno any progress?
AllYourTrees: well theres some rules that require a check of next round
AllYourTrees: idk if you consider that search BrunoFelthes
ddreams: that's still heuristics imo
AllYourTrees: fair
cegprakash: AllYourTrees plz give a replay me vs recurse
AllYourTrees: i linked above cegprakash!
cegprakash: that's u vs recurse
AllYourTrees: oh wait what do you want?
AllYourTrees: ohhh yourself
Westicles: cegprakash whatever joke you are making, we don't get it
cegprakash: yes me vs recurse
cegprakash: choose me in the IDE
AllYourTrees: https://www.codingame.com/replay/557299811
AllYourTrees: cegprakash ^
cegprakash: thx
Queuebee: How did you get that O_o
AllYourTrees: i guess you can play 2 players that aren't yourself in the web ide Queuebee
BrunoFelthes: no struct, i didnt work today
ZarthaxX: gave up bruno?
CouscousAgha: cegprakash you should work on your seeds
BrunoFelthes: sort of it... i was trying to do a MCTS, but the GC killed me
struct: bruno you dont like c++?
ZarthaxX: :(
BrunoFelthes: no... i tried to learn rust, but i m not confortable with it yet
Numby: I had same issue and had to do rediculous things with a pool of nodes and copying into existing nodes
ddreams: it's not such a big step from C# to C++
AllYourTrees: rust MCTS is a pain
Numby: But it's not very good cuz of all the extra crap I put in
kresteodymium: hi guys
kresteodymium: hello world
struct: hi
AllYourTrees: :wave:
kresteodymium: hi struct and AllYourTrees
struct: my ears :scream:
ddreams: pinging with headphones?
struct: yes
CouscousAgha: alright i guess that's it for me
CouscousAgha: this week was a rollercoaster but im proud of my accomplishments
CouscousAgha: see you in the next time boys and gals
CouscousAgha: thank you for hosting this contest!
AllYourTrees: cyas CouscousAgha! nice work!
AllYourTrees: don't forget to read the post mortems CouscousAgha
CouscousAgha: oh where and when should I expect that ? AllYourTrees
ddreams: the forum
AllYourTrees: ppl post their approaches on the forum in the days after it ends
ddreams: most of the top players explain what they did
CouscousAgha: idk if im eligible enough to do one myself
AllYourTrees: oh anyone can, ppl in gold/silver/legend all post
Numby: where do people post
ddreams: even bronze
AllYourTrees: i think they usually email out the forum link
BlitzProg: I'll definitively put one to write how ashamed I am of my performance
struct: any changes I do, do nothing :(
ddreams: I'm following in All_YourTrees' footsteps and copying rules from the cursed
eulerscheZahl: happy 4am
Queuebee: No u
struct: hi euler
AllYourTrees: nice ddreams :D
ZarthaxX: hey toad
CouscousAgha: where's the gold boss at now?
codingWhale: And I thought, you go to bed to a proper time, e... :upside_down:
struct: I wont go to sleep until I get legend
ddreams: you know how old people wake up in the middle of the night?
ZarthaxX: struct but :P
CouscousAgha: struct arent you top 50? lol
AllYourTrees: drink a lot of water before you fall asleep?
struct: im 51th
struct: not good enough
struct: I wanted top 50
eulerscheZahl: i do codingWhale. getting up is another story
struct: oh im 50
struct: bye gn
AllYourTrees: lol!
CouscousAgha: LMAO
codingWhale: Wow, that's impressive. :)
CouscousAgha: well u will get a rerun so dw
ZarthaxX: there will be a rerun struct :D
Queuebee: what's a rerun
struct: Tshirt is so close
struct: yet so far :(
CouscousAgha: u might keep the top 50 as well as lose it!
ZarthaxX: rerun started from actual positions tho right?
eulerscheZahl: in my bed right now. with laptop
struct: yes ZarthaxX
ZarthaxX: struct 30 ranks lmao
struct: euler send help
ZarthaxX: so there is no chance to get top 20 from top 50 structo
Westicles: maybe pb 4 will push you
BlitzProg: I feel tired and a bit sick, but there's 6 hours left
struct: 6 hours is enough time
struct: I could have writen my current bot in that time
ZarthaxX: :rofl:
BlitzProg: And that's what I'm doing
ddreams: rewriting struct's bot?
ddreams: hm
struct: no lol
eulerscheZahl: will this be another BlitzProg story about trying to get legend the last night?
eulerscheZahl: you always write these as your post mortems
struct: euler I even rented aws machine
struct: to test lol
AllYourTrees: :O
BlitzProg: I have very low hopes at this point, but I'm not giving up until I'm absolutely sure I have tried everything
actualCoderTrevor: lol I considered that
eulerscheZahl: but you are a MS guy
eulerscheZahl: no azure?
VincentBab: Stop improving your magic numbers guys, let me have that tshirt please :P
struct: I never used any of them
eulerscheZahl: never
struct: I just picked one randomly
CouscousAgha: I just noticed im the 10th c# player
ddreams: how much did it cost you?
CouscousAgha: do i get anything for that?
BlitzProg: If this heuristic bot doesn't get me back top 200 gold, then I know nothing else was possible for me
struct: the aws machine im using is using windows server so
struct: ddreams no idea lol
struct: pricing calculator was saying 0.5$ an hour
Numby: is it possible to play in IDE vs someone from different league?
Numby: like to test your gold bot vs the legends
BlitzProg: legend can test vs gold, but not the other way around
Westicles: CouscousAgha, you get +50 xp
AllYourTrees: who do you want to play against Numby?
ddreams: ok, so a cheap one
CouscousAgha: Numby you can only test vs top 1000 of your league /and league below u if there are less than 1000 in your league
struct: its 16 threads
ddreams: well, $12 per day
struct: yeah
Numby: I
struct: But I rented i only used it today
Numby: I'd test vs reCurse
struct: on last hours
Numby: :D
CouscousAgha: u want a test vs recurse?
actualCoderTrevor: Numby I've heard you can do it with cgbench but can't confirm
BlitzProg: recurse has like 95% win rate against at least every other players in the entire contest
BlitzProg: it's pretty amazing
**Queuebee just now realised that swapping players swap them in the game and this game is skewed
actualCoderTrevor: Oh wait you said in the IDE
AllYourTrees: Numby you don't appear when i search you for some reason :thinking:
Numby: Hm weird my latest submit is scored at 1260
Numby: gold
CouscousAgha: he must be rank 800+ in gold
CouscousAgha: yeah
CouscousAgha: only top 1000 people are shown for us
CouscousAgha: Westicles good enough for me
nam4.nguyen: Hi
nam4.nguyen: First time for this kind of contest. and I got top 500 but I feel I was so lucky to reach that
struct: grats
AllYourTrees: gz!
nam4.nguyen: Thank you but I wonder about my bot
Counterbalance: nooooo.. 0.18 pts under boss
struct: counter just wait
struct: you will get pushed
struct: I hope
Counterbalance: me too :)
Counterbalance: i'm out of params to tweak
struct: the machine run 300k tweaking params
struct: didnt change anything
struct: ...
nam4.nguyen: I didn't know anything you guys talk about like brutal-tester or beam search etc
ZarthaxX: Counterbalance ebicochineal will push you :)
Counterbalance: yeak i'm watching it :)
struct: brutal tester is a tool to run games localy vs versions of your other bots
struct: beam search is an algorithm
ZarthaxX: oh no
Counterbalance: oh no
ZarthaxX: he beated you
ZarthaxX: D:
ZarthaxX: ah you will prob get there
ZarthaxX: 2 guys submitting
ZarthaxX: you got a win!
Counterbalance: down to 4 now :(
Counterbalance: fingers crossed..
nam4.nguyen: Do you have any docs or something for newbies like me
ZarthaxX: damn ;(
Counterbalance: i'll just resubmit when he's promoted :D
ddreams: nam, you can find cg-brutaltester on github
ddreams: it's a bit of work to set up
nam4.nguyen: Thank you I will try it in next contest
nam4.nguyen: I'm familiar with CP but it's way more diff to this contest
Queuebee: Good night everyone. Good luck!
ddreams: gn queenbee
nam4.nguyen: no best solution, just optimal and try :((
Lysk: just removed opponent sim and boom ranks goes up
Kellthazar: Where I can get the referee jar for the spring2021? I tried to build from source code but didnt work
ddreams: https://github.com/LSmith-Zenoscave/SpringChallenge2021
Lysk: or https://github.com/dreignier/SpringChallenge2021
Lysk: zeno built it though
Kellthazar: Oh, thank you folks
JohnCM: how long do i need to wait to be able to play again if i have an error message
JohnCM: i think i did self-test for too many times
struct: i hate when first matches are 100% winrate
struct: then my dreams get shatered 16 battles after
JohnCM: i'm achieving 80% win rate now in gold
JohnCM: but still cannot beat the boss
ddreams: re_curse resubmitted
Kellthazar: brutaltester worked... nice!
Kellthazar: Released the Kraken
Chainman: :sweat_smile:
struct: im getting tired :(
Chainman: same only few hours left
struct: what time is it there?
Chainman: usa it is 9:46PM in Central Time Zone
struct: damn 03:47 am here :(
Chainman: rough, I know I'm about to lose mental ability soon.
Chainman: It exponentially decreases at a time.
BlitzProg: 4:48am here
Counterbalance: hmm
Counterbalance: i thought my clock was off but it was just silent for 7 minutes :)
eulerscheZahl: :boom:
eulerscheZahl: better?
Counterbalance: :bird:
ddreams: the toad is gone
struct: I think im going to rewrite eval
eulerscheZahl: now?
Wontonimo: 5 hours, that's most of a business day
ddreams: it's painful to recalibrate it
ddreams: but yeah, I'm doing it too
struct: euler I have no ideas
struct: :(
struct: also 5 hours is plenty of time
struct: my current eval is not great
JohnCM: i accidentally shrunk my code in the ide, anyone knows how to expand it
ddreams: ctrl-scrollwheel
eulerscheZahl: the > and v on the left, next to the lines
eulerscheZahl: uncollapse it
Chainman: the touchpad by moving fingers away from each other.
ddreams: I thought he meant zooming
JohnCM: which ones?
JohnCM: yea i zoomed mine to tiny font
JohnCM: i want to unzoom it
ddreams: ctrl-scrollwheel
eulerscheZahl: ctrl+p, type zoom
ddreams: over the text
Chainman: try alt+f4?
ddreams: kick
ddreams: :p
Wontonimo: command + scroll wheel if you are on mac
Chainman: :sweat_smile:
eulerscheZahl: delete your cookies and reopen the site
JohnCM: ah ok euler's method works
JohnCM: i zoomed in to a decent size
Counterbalance: control-+
JohnCM: for a while i count see 100 lines at once
JohnCM: could*
ddreams: well, so far my new eval has a 3.13% win rate
Chainman: I think that's different control +
ddreams: I zoom it tiny on purpose, since I copy/paste it from my own editor
JohnCM: yea i use mac, but i don't have a mouse
Chainman: that's the entire webpage
JohnCM: i think it was probably cmd + mouse movement that caused it, though i couldn't replicate
Wontonimo: command + 2 finger swipe up or down
ddreams: 2 finger swipe == scrollwheel
Counterbalance: move closer to the screen, or further away
Wontonimo: ^^
Chainman: hmmm
JohnCM: oh gosh finally won gold boss
https://www.codingame.com/share-replay/557333095
eulerscheZahl: buy a bigger screen
ddreams: gz
ddreams: now get above him
ddreams: *it
JohnCM: haha but the league is really hard to win
Wontonimo: well, the last 8 hours have not been kind to me for coding, so no more submits for me
JohnCM: ok i've got this, around 70% wins of the gold boss after mass tweaking of heuristics
struct: ill only submit if I get a 65%+ version
Wontonimo: first official contest, and i have to say i enjoyed it
JohnCM: oh won my first 12 arena games, let's see where this goes
BrunoFelthes: welcome Wontonimo
Wontonimo: seeing everyone online at once has been great
JohnCM: hi Wontonimo, thanks for your tips
xoarianuh: I am struggling with this Ghost in the Cell game. What am I supposed to do to produce new cyborgs?
Wontonimo: hey JohnCM , no prob! Did they work?
eulerscheZahl: for his finger tips? are you the Mafia?
JohnCM: the seeding technique works.
i also added in my own move order arrangement tactic to prioritize low cost moves first
JohnCM: save more sun points to react to opponent's growing of trees
Wontonimo: wow! look at you go!
cegprakash: JohnCM what seeding technique
JohnCM: oh doing good in league, won first 17 games
JohnCM: haha Wontonimo, you ok if i share the seeding technique?
eulerscheZahl: low cost first? shouldnt it be high cost?
Wontonimo: go for it
Chainman: oh
eulerscheZahl: grow from 2 to 3, then you have less of size 2
eulerscheZahl: so 1=> 2 will get cheaper
JohnCM: yea so basically knights move away
and you should seed 1 seed each move as it is 0 cost.
grow the seed each round so the seeding move will be 0 cost
ZarthaxX: i think all the people do that
ZarthaxX: :P
JohnCM: yes euler, the growing of seeds to trees are prioritized first
eulerscheZahl: no ZarthaxX, that's pro tactics
JohnCM: but i'm saying i like do actions like harvest, seeding
Chainman: lol
JohnCM: before grow
kovi: morning
struct: I dont do that
JohnCM: so like grow is last action to take each round
Chainman: I've been saying that this all today
BrunoFelthes: wow, morning kovi
struct: hi kov i
Chainman: hello
BrunoFelthes: but i think that people here do not sleep yet :P
AlexiusCzar: how to go back to problem after submtng solution??
struct: I wont sleep until its over
Wontonimo: it's over for me
ddreams: I'm resubmitting my 11 hour old code...
BrunoFelthes: will you not work tomorrow struct?
eulerscheZahl: today*
struct: no, im a student
BrunoFelthes: Oh, nice
ddreams: at least I will beat Ille :p
eulerscheZahl: 4:10 am for him
nam4.nguyen: I'm struggling at rank 300 gold league, you guys have any suggestions? And I wonder when I should haverst trees
JohnCM: my tree harvesting action is tiered based on nutrients remaining
ddreams: you have to post an interesting replay, nam
JohnCM: so if i see opp past a threshold nutrient, i harvest mine too
Wontonimo: hey nam4.nguyen , i'll take a look at your replays ... give me a sec
BrunoFelthes: I was considering take vacations... Im too addicted to work and play..
JohnCM: ohoh rank 30 gold league
eulerscheZahl: i did that BrunoFelthes
JohnCM: haha though i doubt i can win boss, my code doesn't win all the time
eulerscheZahl: and then mostly prorastinated
nam4.nguyen: Thank you guys
ddreams: there euler, I pushed you a bit
BrunoFelthes: ahuahhu, here in Brazil we can have 1 week vacation... but my wife didnt like the idea :P
Chainman: oooh finally I lose to gold boss again,
Chainman: wow
JohnCM: you mean you were winning?
BrunoFelthes: did you liked this game eulerscheZahl?
Chainman: nope, but still losing
Chainman: https://www.codingame.com/replay/557337518
Chainman: I feel I'm using sun points quite efficiently now.
nam4.nguyen: https://www.codingame.com/replay/557332292
nam4.nguyen: I'm losing boss almost the time
eulerscheZahl: board games are never my favorite
eulerscheZahl: it's hard for me as a human to understand what's going on
ddreams: you may consider to only have one seed on the board, nam
JohnCM: i think i'm just gonna sit this one and see where my bot pushes me to. it is around top 30 gold now
eulerscheZahl: so it mostly feels like throwing CPU at it and testing random things to see if it's any better
Wontonimo: hey nam4.nguyen i can't see anything glaring
JohnCM: perfect info board games can use monte carlo tree search
struct: exactly what hapened to me on this contest euler
ZarthaxX: same
Wontonimo: nam4.nguyen i did see in https://www.codingame.com/replay/557336843 that you finished a few s3 trees too soon and by frame 81 your production is a bit low
struct: even during testing
Chainman: I need one more push
Chainman: seems your bot beats mine nam4.
ddreams: I don't think that looks too low at 81, wonto
eulerscheZahl: JohnCM yes, but the state explosion is enormous. that makes it hard for MCTS
BrunoFelthes: but eulerscheZahl, do you prefer games with physics, like meanmax?
JohnCM: what method did you use euler
JohnCM: i mean if you're ok to share
eulerscheZahl: i like grid games like pacman
JohnCM: oh yes i like pacman
eulerscheZahl: i don't like fog of war
JohnCM: rank 23 now:) wonder if this push can make it
Wontonimo: ddreams - you are correct ... lots of S2 trees
BlitzProg: Ah, that feel of resignation, telling me it's finally time to give up
eulerscheZahl: what's the boss score?
BrunoFelthes: i like games with physics, but i think that the lasts with physics do not had too much success...
BlitzProg: I'll try to push some of my old PHP AIs to see which are best, then I'm off
Chainman: 81 moves in the game.
JustAProgrammer01234: some cool song I discovered
https://www.youtube.com/watch?v=dQw4w9WgXcQ
Chainman: Is it best to complete s3 trees when you have 3 or 4 of them on board?
JohnCM: for me, never have more than 4 big trees
JohnCM: cause it will increase the cost of lvl 3 grow too much
Chainman: okay same
JustAProgrammer01234: did anyone check out the cool song I discovered
eulerscheZahl: no
JustAProgrammer01234: ok then
JustAProgrammer01234: here
https://www.youtube.com/watch?v=dQw4w9WgXcQ
ddreams: no, stop spamming tho
JustAProgrammer01234: Im not even spammin
JustAProgrammer01234: bruh
ddreams: also rick rolling is incredibly old by now
ddreams: so try to make it fresh
JustAProgrammer01234: nah
JustAProgrammer01234: still funny
JustAProgrammer01234: you clicked the link didn't you
BlitzProg: ends with XcQ
BlitzProg: ends with me not clicking on it
JustAProgrammer01234: :(
JustAProgrammer01234: ):
JustAProgrammer01234: way better
ClockSort: every internet veteran knows dQw4w9WgXcQ
eulerscheZahl: i just know that i shouldn't click when it's praised as music
JohnCM: well to be fair it's a nice song
ClockSort: i know lol, i always listen :D
KnightMoves: spent the last few hours watching reCurse replays and tweaking my eval function. Finally made it to legend
ddreams: yeah, 99% of the time when someone posts actual music they like it's not something I like anyway
ClockSort: i had to revert back 8 hours of dev time. I was making my bot worse :P
ddreams: really just a waste of time
eulerscheZahl: 100% for me. i don't like musics
ClockSort: ddreams I know a song you would like but i won't post it
ddreams: thanks :)
codingWhale: Nice, KnightMoves! But why did you leave me behind? :expressionless:
ClockSort: i even fixed bugs and that made my bot worse too
eulerscheZahl: i'm the kind of person who skips a 30s movie scene because it's just playing music
ddreams: same clocksort
ddreams: I reverted 11 hours
ddreams: at least that bot got into top100
ZarthaxX: ClockSort first time legend?
ClockSort: my mcts uses a buggy UCT function because that works better than tracking visits properly, haha
ZarthaxX: o.o
ClockSort: @ZarthaxX yes and rank 21! I'm stoked
ddreams: how is it buggy?
ZarthaxX: dude you smashed it gz!
KnightMoves: codingWhale you're so close!
ddreams: maybe you've innovated
ClockSort: it uses global visit count on every node
ZarthaxX: potential t shirt
JohnCM: clocksort your monte carlo is working very well
ClockSort: tracking node visits is extra computation because i store visits on the edges
Wontonimo: global visits?
Wontonimo: instead of parent?
JohnCM: how do you store your mcts search between turns though
JohnCM: is there global memory
ClockSort: yes, the visits of the root node every time :sweat_smile:
codingWhale: I'll just go to bed and hope someone pushes me over. :)
Wontonimo: cheers to you for making that work!
eulerscheZahl: ok, good morning codingWhale
JohnCM: i can't test with my ide anymore, so i'm just testing with arena now
codingWhale: yeah, good night
ClockSort: i don't think i'll make any changes in these last 4 hours
ClockSort: i don't know how to improve
JohnCM: haha clocksort you did well this competition alr
ClockSort: i guess optimising mcts for speed would help
JohnCM: but is there global memory where you can store variables between turns?
ddreams: you don't even use bitboards, right?
Wontonimo: how many simulations are you doing per 100ms ClockSort?
ZarthaxX: bitboarding :D
ClockSort: yes. static State STATES[800000];
JohnCM: i guess any variable outside of the while loop for the game is a global variable?
ClockSort: @wontonimo 12000 rollouts on avg from opening, but on later turns i reuse the tree if possible
Wontonimo: sigh, i'm at 3000 ... too many performance issues
ClockSort: @ddreams that's right, it's an array of 37 ints. horrible :laughing:
JohnCM: yea you probably also prune quite a lot of moves that don't make sense, like one space seeding
Wontonimo: and i've run out of car
Wontonimo: care
eulerscheZahl: others run out of gas
Wontonimo: lol
eulerscheZahl: but not Wontonimo. he runs out of cars
Wontonimo: either I'm balling, or a car thief
JohnCM: codingwhale almost there for legend
eulerscheZahl: whale is gone
Counterbalance: it's blinking
MADDY-1: hii guys
ZarthaxX: annoying blink
Chainman: I can't think anymore, will need to take a nap soon.
Chainman: I think I've been making my bot worse.
ddreams: yeah, I'm off too
ddreams: I'm ok with top 100 this time :p
Chainman: nice,
BrunoFelthes: hope that someone explain how to deal with bitboards at this contest, my bitboard was worsts than the regular class...
ZarthaxX: haha
ZarthaxX: :P
eulerscheZahl: struct has a nice representation
BrunoFelthes: he exposed it?
Chainman: struct has shared?
Chainman: nani?
eulerscheZahl: i just know that he can do the shadow without unsetting bits
struct: o.o
Chainman: hmmmmmm
eulerscheZahl: will you write about it after the contest?
BrunoFelthes: struct, please share it :D
struct: I can sure
ddreams: why would you unset bits for shadow?
ZarthaxX: best bitboard representation ever
ZarthaxX: :)
ClockSort: struct is da man
ZarthaxX: he meant filter somebits in the process ddreams
Chainman: is bitboarding just a mask for where your trees and opponent are? Also I use mask for dormant trees.
eulerscheZahl: not to wrap around. leaving the map on one side and going back in on the other side
struct: BrunoFelthes ill share it after contest ends
ZarthaxX: because calculating shadows requires shifts and depending on your representation wraparound is a problem
ClockSort: you can use a mask to find shadows without a loop
BrunoFelthes: sure
ClockSort: i haven't figured out how though :D
ddreams: hmm.. you can do the 3-shadow without once filtering, struct?
struct: yes
cegprakash: An error occurred (#407): "You reached the limit of plays for a period of time.".
kovi: if you simulate all, it will be O(n) dominated by seeding, so you may not need bits.
storage is 37bytes vs. 4xuint64
BrunoFelthes: oh, i can see now... , one mask for each position
ZarthaxX: position?
ddreams: hmm.. maybe I'm filtering for no reason
ZarthaxX: size of trees
Counterbalance: shadows[angle][size] & 1ULL<<cellidx ?
eulerscheZahl: 1-hot encoding for the trees
struct: I dont use masks
struct: I use shifts
struct: to get seeds I do use masks
struct: but thats about it
eulerscheZahl: just that you use your own cell index and not the one given by the referee
Chainman: that's it I'm out, I still can't beat gold boss :(
ZarthaxX: Chainman gg
ddreams: yeah, I shift too, but I filter once against the bitmask you call "inside"
ddreams: perhaps it's unecessary
SterlingA: Do you guy know any heuristic bot in legend ?
