NitMpez: even having the bonus wrong... didn't make much of a difference... pretty much same rank :P
guspihl: http://chat.codingame.com/pastebin/8eabf09e-7fcc-4650-b870-1c12166e78f2
guspihl: can someone figure out this pattern?
guspihl: its from a reverse mode in clash of code that I didnt solve :(
GoogleJeff: hmmmm. i'm stumped.
GoogleJeff: this is why i hate reverse mode.
GoogleJeff: there should be at least 10 examples for a two-input problem *that* cryptic
ClockSort: now i am curious. looking...
GoogleJeff: oh! i figured it out i think
GoogleJeff: checking...
guspihl: oh sweeet! hope you got it :D can't go to sleep before someone figures this out
GoogleJeff: int((a*a+b*b)**0.5)
GoogleJeff: the (5, 12, 13) is a pythagorean triple.
Tyir: ^
ClockSort: is it just a bitwise operation?
guspihl: ohhhhhhh!!
GoogleJeff: square root of a squared plus b squared.
guspihl: thanks a ton lol
GoogleJeff: rounded down.
GoogleJeff: no worries :)
GoogleJeff: happy to help
guspihl: that's quite interesting actually
ClockSort: sheesh, that's difficult for most people
ClockSort: such as... me :D
guspihl: but it should have given better clues as you said
guspihl: alright, now i can sleep in peace! thanks again, and gn :blush:
NinjaDoggy: FINALLY! A submit that actually gave improvement !!!
miszu: how much improvement?
phord: Got to Bronze in Python and got pretty stuck. Started over in C++. First submission got me to silver.
miszu: congrats!
NinjaDoggy: 15th gold, 71% done
struct: C++ is free legend
NinjaDoggy: :(
Icebox: "free" :thinking:
miszu: meanwhile I almost done with getting all possible transitions
struct: Sorry 20 hours with no sleep
phord: Shit -- the announcement reset my IDE. :-(
NinjaDoggy: also WOAH!
NinjaDoggy: Recurse came out of his rock!
NinjaDoggy: #1 :O
3N0n7: a
struct: annoucement resets ide?
miszu: wow
phord: I was debugging a problem where I sit and REST when I get to zero inventory.
phord: I guess I should have saved the seed.
miszu: and done. Wrote all possible transitions
miszu: but I am worried with pointers and copy by value thingy
miszu: how does it looks like?
miszu: http://chat.codingame.com/pastebin/fd7bdf3d-7000-45bf-a825-fe111681a120
Icebox: looks slow :D
miszu: :(
yhyoxx: :'(
miszu: will it bottleneck for gold?
Icebox: you're lucky that speed doesn't mean anything
Icebox: no obviously not
miszu: ok that's good enough
struct: If php can make it to gold
struct: anything can
rockstar555: how to solve Coder strike Back in Java please help
struct: Its not solvable
rockstar555: sorry not understand
PatrickMcGinnisII: rockstar555 it's a multi, not a puzzle...do better than everyone else
miszu: does this do copy constructor?
miszu: Player p = players[playerPos];
Icebox: yes
Icebox: assuming players is full of "Player" objects
miszu: player is a struct
rockstar555: okk
miszu: yeah players of size 2
PatrickMcGinnisII: struct some of us using php still trying to get to gold *sigh*
Icebox: but an easy way to think about it miszu
Icebox: everywhere where you use "="
struct: Patrick same applies to other languages
Icebox: it's copy
struct: I see c++ in legend and I doubt ill get there
miszu: Icebox so it's better to use pointer?
Icebox: depends
Icebox: it's not alawys better
Icebox: always*
Icebox: but if you have a huge object that you don't want to copy then yes
miszu: well technically I could make a const
miszu: not modifying the state
PatrickMcGinnisII: PY coders are getting better
miszu: ok time to test if getting all acitons is good
miszu: it looks good on first shot
yhyoxx: gj
miszu: Icebox laughing with his O(1) move
miszu: ty yhyoxx
miszu: found a bug :/ multiple casts doesn't detect properly
miszu: nvm it wasn't castable :D
yhyoxx: u should stream
miszu: why?
miszu: I would look pathetic with my c++ knowledge
yhyoxx: not everybody streams to show their skills
yhyoxx: u can stream to have fun
miszu: yeah but what knowledge would I bring to people
Donotalo: found a player who always goes for recipe 0 (as long as there is +3 bonus) and s/he is in top 100
miszu: question for c++: I have a function that returns a vector<pair<Action, int>>. But since it returns by value, it does a copy so it is expensive in this case. How can I return the reference? I am reading you cannot as it gets destroyed in the stack. But isn't there a special rule using the const keyword?
jacek: happy Caturday
Michael_Howard: Test test is chat working again?
__atomic__: nope
Michael_Howard: boo. Better go home then.
Michael_Howard: Stuff appeared again after refresh but I have a gap from 5:10 - 9:31. Is yours the same?
__atomic__: Yes I see the same
Michael_Howard: Cool. If it's not just being random, hopefully it's up again.
Icebox: it is
[CG]Thibaud: pb of certificate renewal
dbf: oh, classic :)
jacek: pb?
eulerscheZahl: problem
eulerscheZahl: (or pb4)
eulerscheZahl: we had that same certificate issue during a previous contest
jacek: is AutomatonNN online?
AutomatonNN: eulerscheZahl is there a way to convert the code to local distance and the next checkpoint is a bit
jacek: he is
eulerscheZahl: not for much longer :rage:
Icebox: AutomatonNN vs Automaton2000 go!
Automaton2000: you dont know the details
AutomatonNN: eulerscheZahl is there a way to see the code in the chat with the leaderboard and i got to legend wi
eulerscheZahl: seriously, stop it!
DomiKo: lol
jacek: yamero!
DomiKo: AutomatonNN one more?
Icebox: when even bots ping euler smh :D
AutomatonNN: eulerscheZahl is there a way to see how the game is working now?
Icebox: :D
eulerscheZahl: AGADE!!!
dbf: he is busy
DomiKo: AutomatonNN lol
AutomatonNN: wow he got interested
DomiKo: :((((
dbf: python is #1 in gold, great
eulerscheZahl: just 0.4 below the boss
dbf: how can you see bot score if you are in legend? :)
eulerscheZahl: and 6.8k players already
eulerscheZahl: open a replay with the boss. cgstats even does that now
eulerscheZahl: http://cgstats.magusgeek.com/app/fall-challenge-2020/betrue12
Michael_Howard: Who coded AutomatonNN and why is he stalking euler?
AutomatonNN: and even if it's not participating with the opposite of the same code
eulerscheZahl: it's agade's bot (which is why i was shouting for him)
eulerscheZahl: something probably went wrong with training the neural network. he has a weird obsession for me
eulerscheZahl: and once called me a "noob"
Icebox: or maybe so many players called you a noob that he just learned it naturally :thinking:
DomiKo: toad is king
Cousnouf: ahah
Icebox: maybe
DomiKo: AutomatonNN what about toad?
AutomatonNN: for real here
Icebox: maybe toad is queen
Icebox: who knows
eulerscheZahl: i have a king in my toad collection https://imgur.com/a/zP9wy6v
Illedan: Good morning
eulerscheZahl: it is. finally something seems to work
Illedan: Nice
DomiKo: Nice one
Illedan: And I see someone came out of hiding -.-
DomiKo: yeah
eulerscheZahl: and an emil deleted the account
DomiKo: and emil. quit ;(
Illedan: Wow
eulerscheZahl: but i think it's not struct
DomiKo: coincidence?
Illedan: That's wierd behavior
eulerscheZahl: weird*
Illedan: Still early!
eulerscheZahl: you make that spelling mistake all the time ;)
Illedan: I know
Illedan: :D
Illedan: thx
DomiKo: will legend reach 100 people?
DomiKo: what you think
eulerscheZahl: i say no
Illedan: Nah
eulerscheZahl: usually the sim-heavy contests have smaller legend leagues
eulerscheZahl: while heuristic games tend to exceed the 100
DomiKo: I done one contest so I'm asking
eulerscheZahl: but will Illedan reach legend?
Illedan: Not sure if I have the motivation
DomiKo: I'm worried about shuffle at final recalculation
eulerscheZahl: i predict neumann will despite him saying he won't
DomiKo: game seem so random sometimes :(
Illedan: Ofc he will
Illedan: Never seen him give up :D
eulerscheZahl: http://cgstats.magusgeek.com/app/fall-challenge-2020/eulerscheZahl
i see no clear pattern here
eulerscheZahl: winning vs higher ranked. losing to those below
rockstar555: hii
Illedan: Not using cgbench euler?
eulerscheZahl: and not enough games in general to deduct anything
DomiKo: In legend we have that weird gaps
eulerscheZahl: nope
DomiKo: cgbench don't give much
DomiKo: you can have 60% winrate with some one
DomiKo: but you play with him only 5 games
rockstar555: hey i cmplete coders strike back
eulerscheZahl: i never used it for more than 10 or 20 games to test
rockstar555: i am in wood 2
eulerscheZahl: the submit button got renamed to "test in arena" for a reason
rockstar555: now i am level 19
rockstar555: hey what is quest in codingame
JFB: Finally Golld. :-)
dbf: and python is in legend, oh
WINWINWIN: JFB are you search based now?
JFB: yes - bfs.
JFB: Still timeouts
JFB: It looks that sort() in C++ sometimes last longer that 20ms
rockstar555: https://www.codingame.com/ide/puzzle/code-a-la-mode
Csipcsirip: now way JFB. how many items in your vector ?
WINWINWIN: Why would that happen?
WINWINWIN: thousands of objects in vector?
JFB: Csipcsirip - sometimes mote that 400k
eulerscheZahl: python in legend. but how?
JFB: eulerscheZahl mayby wrap ?
Csipcsirip: oh okay. than it's possible yea :D
eulerscheZahl: JFB there are server issues affecting everyone
WINWINWIN: Like what recurs did on his bot in CSB?
eulerscheZahl: if it doesn't happen too often, it's not your fault
rockstar555: what was the topic that you are discusing
eulerscheZahl: you get random timeouts with a 40ms limit
eulerscheZahl: contest
rockstar555: which contest
DomiKo: there is only one
rockstar555: fall challange 2020
WINWINWIN: Yes rockstar555
WINWINWIN: Who`s the unnamed player in legend?
WINWINWIN: emil.?
eulerscheZahl: yes
Icebox: rockstar555 do you know how to simulate the game?
DomiKo: euler top10 gz
Westicles: Emil was rust... that narrows it to two options for who's alt it was
DomiKo: any todo to write?
rockstar555: I play played the contest I like contest but when i write code it brew online one
WINWINWIN: He`s not an alt...
Icebox: that was Recurse account
Icebox: xD
rockstar555: i want brew multiple spell to earn money
rockstar555: can any body tell me
JFB: In one step you can do only one brew
rockstar555: Sorry Icebox i didnot simulate the game
WINWINWIN: rockstar555 your question doesnt make much sense.
rockstar555: why
Icebox: rockstar555 how did you solve Vox Codei then?
rockstar555: by using code
Icebox: :thinking:
rockstar555: i didnot understand this
Icebox: you got a lot of "very hard" achievements overnight
Icebox: you must be really good
rockstar555: thanks Icebox
rockstar555: not complete all
rockstar555: 80% completed
WINWINWIN: Nicely done
Icebox: Well you completed Resistance, Music scores, Nintendo (which has 2% success rate!), Cave Dweller, Last Crusade ep 3
LLG: I'm hovering around rank 50 in silver
Icebox: and all of that overnight :o
Icebox: that's some skill
rockstar555: It very hard
LLG: maybe a tiny little imporvement in my heuristics and I can make it to gold
WINWINWIN: LLG I`m in the same boat :)
kovi: 4 more to legend
rockstar555: hii kovi
rockstar555: hey tell me about Code a la Mode
rockstar555: it is three player game
JasperV: same here LLG :D
eulerscheZahl: and a python even
rockstar555: hii eulerscheZahl
JasperV: really cool to have a python submission in legend :)
DomiKo: One python one Java not bad
rockstar555: nice
Westicles: rockstar555, they are gonna get you
JFB: eulerscheZahl - even you in C++ ??? 2020 year is really weird :-)
eulerscheZahl: hii rockstar555
JasperV: A friend of mine is trying to go for it in ruby, that language is not doing well in the rankings :P
blasterpoard: so... I got pushed overnight from #50 to #1 in gold
rockstar555: didn't understand
wlesavo: chokudai is coming :smiley:
rockstar555: Westicles
Westicles: The CG police
rockstar555: what is chokudia
rockstar555: guys one second i will talk to you later
eulerscheZahl: JFB i upped my sim count from 60k to 150k with the language switch
eulerscheZahl: blasterpoard was that #50 before legend opening?
blasterpoard: eulerscheZahl no
rockstar555: https://www.codingame.com/ide/puzzle/code-a-la-mode
blasterpoard: this bot's drafting makes it lose against nnobs, so it doesn't climb
blasterpoard: right now I'm trying to parse 20gb of data I generated overnight to fix that
LLG: I think I can improve my bot by a lot
rockstar555: which type of language like by you eulerscheZahl
LLG: right now I'm only using 5-10ms every turn
eulerscheZahl: even chokudai agrees that legend is hard https://twitter.com/chokudai
LLG: because I prune a lot of states (if they have the same inventory they're considered dupes)
LLG: so I never got past 4-5k sims
eulerscheZahl: rockstar555 why does your profile say that you live in my city, work for my employer and went to my university?
Mohammadmh: hello wrd
JFB: eulerscheZahl - you changed file name from *.CS to *.CPP ;-) ?
eulerscheZahl: and then i hit compile until it did
eulerscheZahl: and then some more till the segfaults were gone
JFB: :-)
DomiKo: then add #pragma and boom
DomiKo: 2x faster
eulerscheZahl: yes, that too
eulerscheZahl: at first i was like: oh, same sim count as before the port :(
eulerscheZahl: then: wait a moment. pragmas!
DomiKo: then: some more pragmas!
eulerscheZahl: copied all from my UTTT at once
Icebox: but then O3
eulerscheZahl: #pragma GCC optimize("-O3", "-ffast-math")
- pragma GCC optimize("inline")
- pragma GCC optimize("omit-frame-pointer")
- pragma GCC optimize("unroll-loops")
eulerscheZahl: anything i'm missing?
DomiKo: then: tricks for inventory
Westicles: avx?
DomiKo: then: every operation in O(1)
Icebox: yes eulerscheZahl
eulerscheZahl: did the bitboard thing in C# already
DomiKo: no need for avx
Icebox: Ofast > O3
DomiKo: you can made every operation in backpack using one operaton
DomiKo: or one + one bit operation
eulerscheZahl: thanks for the Ofast. will use it in my next submit
DomiKo: Ofast isn't faster
DomiKo: its O3 with some more pragmas
DomiKo: you can get worse result with it
Icebox: and O3 is o2 with more pragmas
Icebox: those are optimization pragmas so on average it improves performance
eulerscheZahl: i'm back to watching replays. although it's really hard for me to see what's going on and where my bot makes a bad decision
DomiKo: Euler you can use arch=native tune=native pragma too
JFB: blasterpoard - maybe you have problem with random timeout. And at night, servers run faster, less timeouts and you bot go up
JFB: What is 'arch=native'?
eulerscheZahl: native processor architecture
eulerscheZahl: optimize for that one CPU
JFB: OK, so I add too :-)
eulerscheZahl: not sure if that's smart
eulerscheZahl: CG has different servers
eulerscheZahl: but only compiles your code once
Icebox: generally speaking everything optimization related should be tested
JFB: OK. So it can be problem
Icebox: nothing is "faster" in the world of optimization
JFB: True
eulerscheZahl: i think i should concentrate more on my algo anyways
Icebox: it always depends on architecture, type of task, etvc
eulerscheZahl: 150k isn't perfect but you can build on it
JFB: I think that algo is more important. I usually have > 300 k (sometimes even 700 k) in bfs - and only midle of Gold
eulerscheZahl: BFS, no pruning?
JFB: Not yet
JFB: Next on the TODO list
eulerscheZahl: should i add the urgency bonus to my sim?
eulerscheZahl: pushing that away for days now
DomiKo: I guess yes
JFB: Probably more important before last brew
Westicles: Detect which brew the opponent is going for and prune it from your search?
LLG: what kind of difference does urgency bonus make?
rockstar555: yes i am from furth eulerscheZahl you know this place
LLG: since it's already added into the recipe price anyway, no?
rockstar555: but i am not in your company Sorry
Icebox: why would you put it then as your company :thinking:
rockstar555: I like your company
Nerchio: :joy:
rockstar555: Siemens Energy so I edit this
rockstar555: I like it names
rockstar555: Sorry
Icebox: what a coincidence
Icebox: to live and study where eulerscheZahl studies, and like his company name :o
rockstar555: :sweat_smile:
Csipcsirip: this guy is a big troll
Icebox: yeah and he got 8 Very hard puzzles overnight
Icebox: damn I wish I was think good
rockstar555: what is you saying Csipcsirip
Icebox: this*
Nerchio: maybe it's another recurse smurf :joy:
WINWINWIN: Is it zarthaxx?
WINWINWIN: :D
Icebox: I don't think either of them has enough time for this
WINWINWIN: Yeah...
rockstar555: :innocent:
wlesavo: damn, half day rewrite with no tuning to beam search already at the same level as my previous version
wlesavo: or is it even better
wlesavo: wtf
WINWINWIN: :P
rockstar555: :100:
WINWINWIN: wlesavo can the brutaltester help with figuring out if the rewrite is good?
rockstar555: I am learning programming language for 6 years
WINWINWIN: Sorry, I have never used it... dont know how it works/
rockstar555: no
rockstar555:
WINWINWIN
WINWINWIN: Yes?
rockstar555: how this is possible
WINWINWIN: what?
JasperV: each time I open the chat I see more nonsense :joy:
WINWINWIN: :D
LLG: WINWINWIN I respect you for doing this in python 3
WINWINWIN: Its the language I know the best :P Not out of choice, its necessity
LLG: I wrote my bronze league bot in TypeScript
LLG: completely rewrote my silver bot in C#
WINWINWIN: Even more brave :)
LLG: and I'm even contemplating a rewrite in C++
rockstar555: i like most paython
rockstar555: then i solve puzzle more in paython
eulerscheZahl: sadly paython isn't free :(
i'm a big fan of open source
wlesavo: WINWINWIN it probably can, but i assumed this by submit perfomance
JFB: eulerscheZahl - is not free?
eulerscheZahl: read closely JFB
LLG: wouldn't be called paython if it was free
WINWINWIN: :P
WINWINWIN: Understood wlesav
WINWINWIN: *wlesavo
Icebox: I'm a fan of freethon
Westicles: Is there any way to get at a full dictionary for a puzzle?
rockstar555: what is Triangulation Puzzle
Icebox: the puzzle you solved tonight rockstar555 :)
WINWINWIN: He solved it?
JFB: eulerscheZahl - I'm slow to catch the joke ;-)
Icebox: of course!
LLG: I'm still on probationary period for my company
LLG: and I'm using company time for this
Icebox: and got 100% + 3 remaining jumps on the 8th test for a 750xp achievement WINWINWIN
Icebox: together with 7 other very hard puzzles
LLG: I hope my superiors aren't here xD
rockstar555: ohh you still looking what puzzle is solves by me
Icebox
rockstar555: nice person
Icebox: I don't like cheaters rockstar555
WINWINWIN: I dont get it, is triangulation shadows of the knight?
WINWINWIN: Which episode?
JFB: LLG it is Saturday - so you probably do not need to do job?
rockstar555: I am not a cheater
rockstar555: I know i solved it
WINWINWIN: Yes
LLG: yeah it's late Saturday for me right now
LLG: I'm not worried about right now
LLG: rather, the entire last week
LLG: xD
WINWINWIN: What is triangulation AutomatonNN
AutomatonNN: etc etc
ParticleBeam: I suppose thats not wrong
Westicles: Nothing in /usr/share/dict
rockstar555: Triangulation facilitates validation of data through cross verification from more than two sources. It tests the consistency of findings obtained through different instruments and increases the chance to control, or at least assess, some of the threats or multiple causes influencing our results
Icebox: https://www.betterevaluation.org/en/evaluation-options/triangulation#:~:text=Triangulation%20facilitates%20validation%20of%20data,multiple%20causes%20influencing%20our%20results.
Icebox: write your source
Icebox: when you quote it
WINWINWIN: Come on man...
WINWINWIN: its like a badly coded NN
Icebox: :D
WINWINWIN: :D
rockstar555: :thumbsdown:
rockstar555: I copied but not from here
I
https://www.betterevaluation.org/en/evaluation-options/triangulation#:~:text=Triangulation%20facilitates%20validation%20of%20data,multiple%20causes%20influencing%20our%20results. I copied from where i learnd it from.
WINWINWIN: Learnt what?
eulerscheZahl: *insert more copied text*
WINWINWIN: :P
DomiKo: https://www.codingame.com/replay/507838097
DomiKo: ohhh nooo
DomiKo: 10 tier3 in backback
DomiKo: and no potion to do :(
DomiKo: enemy sim is so powerfull :(
eulerscheZahl: classic
Velicko: I can't manage to set up brutalTester, get timeout for both players. I've copied CLI class and pom.xml from https://github.com/Akarachudra/brutaltester-spring-challenge-2020. Is there something else i need to do?
LLG: lol the 10x yellow problem
LLG: I rewrote my heuristics like ~20 times before I kinda solved it
LLG: interesting to see someone in Legend still encounters it
wlesavo: problems of whote people :smiley:
wlesavo: white
DomiKo: 10 tier3 is so good bro
eulerscheZahl: i fall for it too
LLG: at least you get 10 bonus points in the endgame
LLG: xD
blasterpoard: this submit looks like adding 30KB of stats to my code is a very useful thing
karliso: Velicko https://github.com/crazyiop/FallChallenge2020
LLG: funny how my bronze league bot has a time complexity problem
Velicko: thx :)
LLG: then my silver bot does anything in 10ms
Westicles: stats like an endgame?
karliso: Velicko timeout could be a problem in your code though
blasterpoard: mostly spell selection
Nerchio: going legend blaster? :D
blasterpoard: no, going to lose to the boss 5x in a row
WINWINWIN: How much behind the gold boss are you?
blasterpoard: im 0.3 ahead
WINWINWIN: Come on, why the no?:)
WINWINWIN: :)
Nicopen: -
blasterpoard: I think I just promoted saelyos
wlesavo: easy push for saelyos
DomiKo: legend is getting big :/
WINWINWIN: 37 isnt big!
blasterpoard: so far, I managed to deflate gold boss's rating by 0.7 or so
wlesavo: blaster will push half of gold to legend
JFB: DomiKo 37/6814 ~ 0.5 %
blasterpoard: wlesavo you're welcome
DomiKo: but blaster is pushing :joy:
wlesavo: ill push boss back up dont worry :smiley:
blasterpoard: it's happenning, a chain of losses at 95%
JFB: finally blasterpoard :-)
Saelyos: oh thank blasterpoard :)
Nerchio: so training vs boss in the ide is fine even though its rust?
blasterpoard: Saelyos np
WINWINWIN: rust doesnt work in the IDE right nerchio?
AntiSquid: RIP email.
AntiSquid: emil. *
Nerchio: i think it works
wlesavo: gratz blasterpoard
AntiSquid: gratz blaster
Icebox: AntiSquid emil is doing quite fine
Icebox: in the 1st place
Icebox: ;)
WINWINWIN: He really is an alt?
Icebox: emil deleted account the moment recurse hit legend
AntiSquid: i know
Icebox: also if you want more trivia
Icebox: https://nierautomata.wiki.fextralife.com/Emil
JFB: It is the pure coincidence ;-)
Icebox: and if you know what the previous icon of him was
blasterpoard: it's not like he was actually hiding his identity, on discord he was quite open about it
AntiSquid: i guessed it before lol days ago
WINWINWIN: Nah, reCurse doesnt need an alt..
AntiSquid: when we were talking about alts
AntiSquid: faint smile avatars both .
Icebox: Or you know... both from nier automata
dbdr: gogogo wlesavo!
AntiSquid: i didn't play the game so no didn't know that Icebox
wlesavo: dbdr coming :slight_smile: half a day rewrite worth it
ntroPi: whoa! thanks for the #pragmas
ntroPi: They literally doubled my search depth
geppoz: c++? which are you using?
**LordSkaven wishes C# could do the pragma thing
wlesavo: yay
ntroPi: yes C++
blasterpoard: gj wlesavo
wlesavo: thx
ntroPi: #pragma GCC optimize("-Ofast", "-ffast-math")
- pragma GCC optimize("inline")
- pragma GCC optimize("omit-frame-pointer")
- pragma GCC optimize("unroll-loops")
geppoz: ty
Westicles: Can't you do mcs -optimize for C#? Or is that on by default
LordSkaven: I think that is on by default when building in release mode
BoBot: omg, while microoptimizing my code, I found a < vs <= error, which made my bot never use more than 9 inventory slots...
BoBot: how the hell did I make it to Gold...
LordSkaven: I just realized that I have the same bug BoBot
Icebox: :clap:
rockstar555: hii
LordSkaven: hey rockstar, how are you doing today?
BoBot: #pragma GCC optimize("unroll-loops")
lol, I unrolled my for in in range(4): in Python, made the relatively simple, but oft-used method go from 20% of time usage to ~13%...30% optimization...would have thought Python can optimize those things automatically at least
rockstar555: nothing
Icebox: nah BoBot
Icebox: not with python
Icebox: we don't do those
Icebox: :)
blasterpoard: now that I'm in legend... time to finally stop ignoring urgency bonuses, I guess
emh: hey.. I need help beating Silver. any tips?
BoBot: yea...I actually saw zapakh do all of those t0, t1, t2, t3 ingredients separately, but I delayed that as long as I could, and tbh, it is not that much of a boost, I am just trying to get as much as I can out of Python now
blasterpoard: emh beam search to highest score + inventory value
blasterpoard: seems to be the standard gold strat
emh: ohh ok
emh: why BFS is not good enough? don't know beam search
emh: will have to learn it then
emh: I guess
blasterpoard: beam search = bfs, but keep only N nodes in every layer
blasterpoard: bfs doesn't have enough depth
blasterpoard: to reach 2+ potions
BoBot: simple BFS was not fast enough for me (Python), the branching factor is pretty high in this game
ntroPi: beam search is linear time complexity, you can go much deeper.
eulerscheZahl: can you sort in O(n)?
blasterpoard: yes, if you use bucket sort
ntroPi: nth_element
blasterpoard: which is surprisingly applicable hear
blasterpoard: *here
emh: ok so what I do is first do a reverse BFS from all potions to all states. that completes fully. then I do another BFS which uses the distance to nearest potion to limit the number of nodes. so maybe it is a beam search already
blasterpoard: nearest potion isn't enough
wlesavo: i had mc for mid gold, but it was not enough for legend
emh: ahh ok not just nearest potion but price / distance to nearest potion actually
emh: or max of that I mean
ntroPi: you will need to take enemy moves into account at some point
emh: damn it.. was hoping could get away without that ntroPi
emh: for now
emh: is it necessary to get to Gold?
ntroPi: what league is for now?
emh: Silver
blasterpoard: you don't need to look at your opponent in silver
BoBot: @emh I pushed to Gold finally, but tuning the endgame strategy - finding "killer moves" that seal the game, also a lot of tome LEARN tuning
blasterpoard: just get an OK search
ntroPi: I neded it for gold but with super efficient brewing you may make it to gold without
wlesavo: win/lose estimation with basically any search should be enough for gold for sure
emh: maybe my search is buggy.. sometimes the opponent get 6 potions while I get 3 or 4. I have a "fastest" mode which sets the brew price to 1 for all brews if I see that I am behind on potions
ntroPi: fastest mode will be bad ... except on last potion
ParticleBeam: Remember opponent might brew your "fastest" brew before you
emh: hmm ok
emh: will try to put it for last potion only then and submit
emh: I've tried with both tomeIndex > 0 enabled/disabled. right now enabled and I'm 157 in Silver
ntroPi: Ok sad thing about the #pragma - all that extra performance reduced my rank by 40 :-(
emh: my "beam" search is expanding around 30k nodes and spends around 20ms doing that
emh: have a priority queue and skips states seen again which have a better path to them
Pavasaris: How to place one tier-0 ingredient upon the spell? :thinking:
emh: Pavasaris automatic when learning
Pavasaris: But the number of tier-0 ingredients is not decreasing while learning.
Icebox: the lower the spell the cheaper it is
Icebox: so the lowest one costs nothing
LordSkaven: the first one is the list is free to learn
Icebox: while the 2nd one cost 1 ingredient
Pavasaris: ok, got it.
Pavasaris: thanks
SabertheLost: So, I increase in rank if I cast a spell that is not part of my algorithm instead of sleeping. I need a better algorithm :)
SabertheLost: (As long as it´s not a base spell)
geppoz: how do you weight price/distance?
geppoz: e.g. better 13rupies in 3 turns or 10 in 2 turns?
geppoz: which factor seems best?
Icebox: just compare it to how fast the opponent can do those potions :thinking:
geppoz: i mean, without opponent
geppoz: im behind :D
ntroPi: 10/2 > 13/3
This assumes that your predictions are exactly correct though.
Icebox: is it though ntroPi?
geppoz: ntroPi why? and 9/2 ? 8/2?
Icebox: well this one's easy
Icebox: :D
Icebox: oh or you mean
Icebox: nvm
Icebox: I though you were asking if 9 in 2 is better than 8 int 2
Icebox: in*
geppoz: i.e. how many rupies worth wait a turn?
yhyoxx: rupees per turn
ntroPi: it depends frwom where you're counting the turns
geppoz: in silver version without calc opponent
geppoz: from current state
ntroPi: from current turn -> hard to say.
from last brew -> 10/2 > 13/3
ntroPi: but there is a lot of extra stuff that can change it. Like enemy actions, game end, next potion ...
ntroPi: @Icebox or am I missing even more?
Icebox: I mean if you look at it in a way
Icebox: like using resources
Icebox: as in spells are your resources
Icebox: what are you left with?
Icebox: maybe 13/3 leaves you with betters spells
Icebox: or more efficient rests
Icebox: also you spend 2 turns progressing anyway
Icebox: so the 3rd action is like
ntroPi: both of which are not clearly measurable either.
Icebox: "would I do X" for 3 rupees
Icebox: if the extra action is resting then 13/3 is always better because you have to do it eventually anyway
geppoz: i have no speed problems, but having hard time to score states
ntroPi: Yea, I think you're right, in the end the search algo must decide and no fixed score can be decided beforehand.
Icebox: sometimes the 13/3 will be because you need to learn an extra spell
ntroPi: geppoz try not giving it a numerical value but compare states in a more complex way.
Icebox: which means you spend a turn learning which also progresses you
Icebox: so it's like "learn this spell and get 3 rupees for free" deal
Icebox: 10/10 would do
ntroPi: I completely agree Icebox - I just interpeted the question as: Assuming everything else is basically equal, which should I prefer.
I'm aware the assumption is shaky at best.
Starbender: how do you guys measure your execution time ? I measured up to 6000ms on my first iteration and do not timeout, that means the execution is interrupted, what now ?
Starbender: I wish codingame would display the time it measures
geppoz: ok, but question then is: how much worth have learnt an extra spell? :D
therealbeef: did you start counting when the first stdin comes in?
geppoz: ty for tips, let me think about
Starbender: @therealbeef oooh
Icebox: geppoz it's worth a lot
therealbeef: waiting for stdin is not part of time i assume
Starbender: I definitely counted the stdin
Icebox: the more spells you have the better
Icebox: at the start of the game you learn spells without any direct game
Icebox: if you can associate gain with a learn action then that's just really good
Icebox: imagine if you had an option between cast A + cast B and learn C + cast C
Icebox: C is always better
Gronahak: if you don't time out because of the extra branching factor :p
Starbender: thanks @therealbeef you helped me out 😁
WINWINWIN: Neumann in legend :) euler was right
therealbeef: :-), this had me puzzled for a while too ;-)
MisterCage: i am new and i want to practice but i am confused where can i find the coding challegnes? ai or algorithms?
Icebox: https://www.codingame.com/contests/fall-challenge-2020
Icebox: MisterCage
MisterCage: huh thanks
wlesavo: wtf, i tweaked a single parameter and got to top6 with smth like 30k node visits
ZarthaxX: 30k node visits? wtf
CommanderCero: thats not much i reach 100.000 and some claim to reach 200k :sweat_smile:
Nerchio: yeah but he's using beamsearch without duplicates i think
Icebox: that's kindof the point that it's not much CommanderCero
Nerchio: you can have 200k nodes and 150k useless
CommanderCero: yep thats the case with my search
wlesavo: ZarthaxX yeah, its unique count
jft63: wlesavo to check the the unicity, it's based on the action list or on the state content?
wlesavo: state, you can use spell and then rest or rest and then use, same actions different results
ZarthaxX: ahh okey
ZarthaxX: beamsearching and 30k uniques wlesavo?
jft63: interesting, thax
jft63: thx*
wlesavo: yeah, suddenly thats enough
ZarthaxX: im amazed that you are beamsearching and you have such good results
ZarthaxX: i thought it was hard not to prunne useful ones
ZarthaxX: honestly, :clap:_
ZarthaxX: gz on that one
Bikoko: when will the laderbroad update?
wlesavo: spend only a half a day on this rewrite also, wasnt even gonna try
AntiSquid: the updates are always live and up to date, not sure what you mean Bikoko
ZarthaxX: half day :rofl: sheez
ZarthaxX: nice change
wlesavo: still not sure why this even works
wlesavo: i also have deprecated mc for opponent prediction :smiley:
jft63: you mean you do not predict opp?
wlesavo: i do but very vvery badly
jft63: ok
AntiSquid: :thinking: so bad enemy prediction is the key !
wlesavo: yeah nobody expects that!
CommanderCero: great i already have that
CommanderCero: wheres my t-shirt
wlesavo: CommanderCero now you just have to reach good players
AllYourMCTS: are there different tshirts for every contest?
wlesavo: i know there are 3 sprites to choose from, so you can get different ones
AntiSquid: robot hand is best one
wlesavo: i choose that one for ooc yeah
yhyoxx: u guys are getting tshirts ? :thinking:
AntiSquid: it's top 20 only
yhyoxx: the only goodies im getting are headaches
yhyoxx: and KNOWLEDGE
AntiSquid: the friends you made along the way are the real headaches
yhyoxx: ???
AntiSquid: it was a joke
AntiSquid: never heard of the meme "the real prize are the friends we made along the way" ?
karliso: "If the absolute final spell is learnt on the same turn as a read-ahead tax is applied, the extra tier-0 ingredient is discarded." -- what does this mean?
eulerscheZahl: when you empty the tome completely
Nerchio: don't bother it doesn't happen :D
AntiSquid: if the last spell that can be learned is learned . the taxcount can't be gained
BoBot: make it happen!
eulerscheZahl: then there is no more spell to place the tax at
karliso: oh
eulerscheZahl: so your taxes goes into the void
BoBot: no resting, only learning and casting!
Starbender: but isn't the last spell free ?
Icebox: not if you both pick with enemy
Icebox: and different spells
eulerscheZahl: if there are 2 left and you take the expensive oneand ..
Starbender: oh right
eulerscheZahl: won't happen in legend games anyways
AntiSquid: or if you play it in 4 players:
https://spicee.mattle.online/lobby/rooms
NojuZ: I keep getting "has not provided 1 lines in time." I even put cout << "BREW" at the very start of the program and it still doesn't recognize that I'm outputting something. What could cause this?
Csipcsirip: try comparing your code with the default one
NojuZ: I don't have much honestly, just a few vectors and structs. I only store the data. Shouldn't affect anything that comes before their declaration.
NojuZ: I put marked everything but cout as a comment and it still gives me the error.
geppoz: cout << "BREW"<<endl you mean?
NojuZ: Yeah.
eulerscheZahl: " BREW " << id << endl
NojuZ: Oh nevermind I was missing endl. Thought I added it
geppoz: (if it was id missing you would have invalid move, not no output)
Bert.Russell: Hello I get random timeout errors in the Arena, but when i choose "send game parameters to IDE" I dont get such errors.
Has anyone encountered sth like that?
MadKnight: because it's random
MadKnight: as u said
geppoz: yes I did in python
Icebox: and
Icebox: I
Icebox: did it
struct: Did you sleep yet?
Icebox: no ofc not
Icebox: imagine sleeping smh
yhyoxx: what's sleep ?
DomiKo: what happend to Rounddice?
ITACIRE: hi everyone
yhyoxx: fen al khawa
yhyoxx: hello*
ITACIRE: I'm doing an exercise in here
yhyoxx: good luck
ITACIRE: I have to return the ma element of an array
ITACIRE: not using functions
ITACIRE: i've tried lot if etc but still
ITACIRE: :grimacing:
ITACIRE: need a hint
yhyoxx: not using functions wym ?
ITACIRE: cause when I use them it doesn't it tells me that i fail
ITACIRE: i'm using php
ITACIRE: and when i use max(array(1,5))
yhyoxx: just google how to get max element of an array
ParticleBeam: Time for my 99th rewrite
ITACIRE: lol okay thanks :wink:
Panagis: Warning: your code did not read all available input before printing an instruction, your outputs will not be synchronized with the game's turns and unexpected behaviour may occur
ParticleBeam: Your code might be timing out
Panagis: how i can see my timing code?
Earthborne: time your code
Panagis: what is the max time per round?
Earthborne: 50 ms
Panagis: ok thanks
Panagis: if i exceed the limit what happends?
Panagis: i am asking because i think this is my problem
reCurse: You get that warning
Earthborne: You get out of sync and the game thinks you're printing before reading input for next round
Panagis: mmmm
Panagis: thank you very much
JFB: Panagis - this message mean that your first output (print) is BEFORE you read all data in the round.
JFB: It can be OK. For example I print debug message when create some data before I start reading. And receive this message.
JFB: It is warning - and as with all warnings - it better to avoid them but sometimes, if you understand the reason, is OK to have it
meh1001: If you use file=sys.stderr in your debug prints it won't give you the warning
meh1001: err, sorry I guess that only appluies to python. If you print your debugs to the stderr stream, however you do tha tin your language
Earthborne: Here it's most probably a symptome of being too late from the round before, though
Counterbalance: I don't get that warning and I also print to stderr before reading any input
JFB: As you aid it I checkae and and now I do not have this nmessage. Strange, 2 days ago I'm sure that I have had it
thibpat: hey all, I'm streaming my progress on the Fall challenge (javascript / english), trying to climb the silver league, feel free to join https://twitch.tv/thibpat :wave:
eulerscheZahl: DomiKo 03:33PM
what happend to Rounddice?
he lost his corners?
Icebox: :|
eulerscheZahl: be careful Icebox or you will share that fate
AntiSquid: added 1 line to my wrapped 10 liner ruby code and suddenly getting decent winrate
reCurse: Lost his edge
AntiSquid: sorry for offtopic, Rounddice best dice
JFB: AntiSquid - add more lines to promote to Legend ;-)
DomiKo: corners maybe
DomiKo: but points for sure
JFB: AntiSquid - I do not understand - you use C++ to run Ruby code ??? (I see that you use C++ and you write about wrapped Ruby code)
mlemm: can i play my code on CG from multiple source files ? it's getting very messy here with +700 lines on one single file :disappointed_relieved:
struct: Use local ide and merge files
struct: There is an extension to sync one file with cg
AntiSquid: submit the binary
struct: and get disqualified
struct: good advice
mlemm: wow really
mlemm: thanks struct ill try that
struct: https://www.codingame.com/forum/t/codingame-sync-beta/614
eulerscheZahl: https://www.codingame.com/forum/t/cg-local/10359
mlemm: :pray:
eulerscheZahl: stupid bot refuses to win :(
AntiSquid: 6849
AntiSquid: participants
**eulerscheZahl predicted 7k on last saturday already
struct: euler was right
struct: Impresive numbers
AntiSquid: forgot to deduct email.
struct: in comparison with previous contests
CommanderCero: can you also predict my lotto numbers pls
AntiSquid: the ones you pick, not the winning numbers then :P
wlesavo: wow that submit from agade
JFB: wow
struct: not bad
Donotalo: what strategy will take to legend from gold?
AntiSquid: winning
Icebox: damn
wlesavo: AntiSquid dont be so harsh, not loosing is enough
Icebox: those are some wise words
Donotalo: :( any better word so amateurs understand?
therealbeef: if you dont play, you cannot lose
BoBot: wow, Gold boss learned 13 spells in one game
hasleron: Do you guys know these "bearing bugs" where you hopefully fix them just to lose 300 ranking places as a result? #fml
BoBot: 39 turn game, beat me with 101 points
https://www.codingame.com/replay/508086303
JFB: In Legend only C/C++/Rust (+ 1 Python & 1 Java) It looks that it is really CPU contest
BrunoFelthes: 2 java
BrunoFelthes: it is not CPU... I'm checking 10k states
meh1001: I can only check 650 nodes reliably without timing out with my current crap pile
jft63: you are just checking the right nodes :)
BrunoFelthes: that is the thing @jft63
jft63: BrunoFelthes are you simulating opp?
BrunoFelthes: yes
BrunoFelthes: 6ms for op, 30 for me
jft63: ok
BrunoFelthes: 14 for GC
JFB: Sorry BrunoFelthes - I missed 2nd java. Of course if you are really smart you can check only 1 node - the best one :-)
BrunoFelthes: I'm just saying that you can get legend checking 5-10k nodes...
BrunoFelthes: I did it
BrunoFelthes: And I did it without local tuning...
BrunoFelthes: My last step is build a local simulator to tuning my magic numbers...
ddreams: there's usually publicly available ones
Westicles: Do you guys think a scrabble optim would be fun? You get results that look like this
BrunoFelthes: yes, i'm building a GA algo on top of the brutaltester
Westicles: http://chat.codingame.com/pastebin/0ec36cab-c613-4a66-a31b-ce8be0cbc36c
wlesavo: im honestly dying to read Fangel's pm
lowzhao: i still not sure what is the criteria to prune the branches do you guys have any idea?
jrke: wlesavo what you did to be in legend
wlesavo: jrke rewrited to beam search yeasterday
JFB: BrunoFelthes - in topcoder is person (wleite nickname) from Brazil to, who always use Java. And often win with C/C++ programs.
wlesavo: and fixed most of the bugs today :smiley:
BoBot: @wlesavo because Fangel is Python?
jrke: wlesavo beam width and whats the depth and nodes visited ?
wlesavo: jrke somewhat low, like 30k unique visits
jrke: mine unsubmitted beam search is very much unoptimized rarely touches 2k
wlesavo: didn't even have to tune, second working submit got to legend
wlesavo: i had some low counts because of a lot of bugs at the start also
m0u53z: ok
wlesavo: but i believe 30k is still very low
lowzhao: wlesavo did you used any algorithm for learning?
jrke: wlesavo beam width?
wlesavo: well learning is incorporated in sim lowzhao
wlesavo: jrke 500-700
wlesavo: not a lot also
lowzhao: thank you!
geppoz: depth?
jrke: wlesavo are you using vectors or anything else?
Aravindhsiva: https://www.codingame.com/clashofcode/clash/14723200d5c597610b6a894e48b35c55bee05cb
Aravindhsiva: Interested can join
AntiSquid: Aravindhsiva https://www.codingame.com/contests/fall-challenge-2020/leaderboard/global
wlesavo: jrke vectors and pq
jrke: how do you sort vectors?
Westicles: What is a smurf (in CG context)?
Astrobytes: alt account
Westicles: ah
Beresta: with beam search, do I understand correctly you use some heuristics to evaluate the state and keep N best ones to look further?
Csipcsirip: ye you give an eval score to every state and keep N best after every depth
struct: From what i understand its bfs with pruning
Beresta: how do you incorporate enemy moves for beam? I guess if they are part of the search - you will just look for worst enemy moves and best your ones :D unless heuristics maybe balances both somehow, or enemy is just handled separately
Icebox: looking for enemy worst would find you the best choices if enemy is trolling
ClosetAI: With enemy present it's much harder to evaluate states and also number of states is squared
struct: yeah, if you have 6 moves and enemy has 7 it will be 42 combinations on depth 1
jrke: how you guys sort vectors queues or arrays in c++ for beam search?
CommanderCero: std::sort
jrke: it takes lot of time for me
Beresta: I guess some sort of priority queue (heap or something)
CommanderCero: although i use a heap, but people didnt like
CommanderCero: that
ClosetAI: nth_element should be better in theory but it shouldn't matter really
CommanderCero: isnt that essentially partially sorting
ClosetAI: yes
jrke: can anyone provide me link for heap in pq
CommanderCero: what do you mean with pq
lifetimeLearner007: Hi. I'm using Javascript for this contest. seems 'readline' is very slow and taking most of the allotted time. Can anyone suggest a faster alternative?
Earthborne: Clock starts after input is given :)
Beresta: I guess that? https://en.cppreference.com/w/cpp/container/priority_queue
wlesavo: yes
Beresta: funny that apparently there is no default implementation in C# somehow
lowzhao: why gold boss become unpredictable?
lowzhao: some game with different outcome
CommanderCero: timing issues
CommanderCero: probably
lowzhao: I see.
lifetimeLearner007: yes @Earthborne, but reading the input and storing it shouldn't take much time right. But this is taking too much time.
eulerscheZahl: are submits getting slow again?
Earthborne: Are you timing from after readline?
CommanderCero: yes
CommanderCero: otherwise you will measure the time it takes to get the input
CommanderCero: including any delays
Earthborne: ^
Csipcsirip: any tried calculating some stuff on a thread while reading inputs btw ?
lifetimeLearner007: yeah logging time taken around readline
wlesavo: here is some top20 bot gameplay https://www.codingame.com/share-replay/508140389
lifetimeLearner007: const t = new Date().getTime();
const actionCount = parseInt(readline()); // the number of spells and recipes in play
console.error(new Date().getTime() - t);
This is taking 30+ ms
tobk: I don't get it. My bot is occasionally timeouting, but the only bit that could timeout is running in a timed loop with some very generous buffer (just 30 of the 50 ms)
wlesavo: tobk same, had to go to 30msand still some timeouts
Earthborne: That shouldn't matter on you time budget lifetimeLearner007
try: inputs -> start clock -> calculate -> stop clock -> print
tobk: lifetimelearner if you start the timer before taking the first input, you effectively count both your and your opponents turn
Batpapa: Hi! Quick question :) In C++, is the arena compiled in release mode? Or in debug like in the IDE?
Zenoscave: wlesavo why? End search?
lifetimeLearner007: oh! is it?
I didn't know this. Thanks @Earthborne and @tobk
reCurse: Depends on the language
struct: Batpapa pass O3 flag
Batpapa: how??
struct: #pragma GCC optimize("O3","unroll-loops","omit-frame-pointer","inline")
struct: at the top
struct: You dont neeed all of that
wlesavo: Zenoscave i think weight of learning was bigger then the discount of winning one turn later, smth like that
Batpapa: wtf thx!!
struct: O3 makes it harder too debug though
Batpapa: sure
Batpapa: but just for the arena, not usually ;)
JFB: wlesavo, tobk - it my bot timeouts generate sort() (standard C++ sort). I have had sometimes 200k elements to sort - and sometimes it last 20 ms. So even with 30 ms time limit I have had timeouts.
tobk: My code is, very roughly, read-inputs, start timer, search-loop to find best action for 30 ms, if no result get any applicable action (e.g. just learn any spell or rest), and still sometimes timeout :-/
wlesavo: oh no, not that. search shouldve ended earlier, it means idk why
Zenoscave: ah
tobk: I have some debug output after the 30 ms, but neither that nor the fallback-action selection should take 20 ms (and most of time time it takes only ~2ms according to anoter timer-output after that)
JFB: tobk - I do not know your bot - but in my it was problem with sort() of big array. I spend few hours becouse it was only one, "innocent" line
tobk: JFB Thanks, but I only keep track of the single best action (actually, best path, but those are never very long, seldom more than 20 elements)
Batpapa: Alllllright! O3 flag just made me explore twice the number of nodes :o
Batpapa: Big tank you!
Washier: <jealous C# guy iver here>
JFB: OK tobk. So it is not your case.
struct: C# has release mode
struct: cant be jealous
Washier: <shocked pickachu meme>
Counterbalance: tobk are you printing the total time right before you print the action? In the 'main'?
Washier: how?
struct: its on release, you dont need to do anything
Washier: oh, ok good to know thanks.
eulerscheZahl: my sim count is zigzagging like crazy. 150k in one turn and 40k in the next. :/
tobk: Counterbalance I take time twice: once in the main loop, and in my "search-action" function. the 30ms timeout is in the search-function. I print both the search-time and the total time.
Counterbalance: and the total time is <= 45ms?
Zenoscave: Same eulerscheZahl
eulerscheZahl: then i blame the servers
Zenoscave: My code is fairly inefficient. not sure if it's servers
Counterbalance: I had timeouts like that too; debugging the time + action, next line would print the action (in main). It would timeout, even in ide.
Zenoscave: GC is still my issue
JFB: eulerscheZahl - I think that some "black hats" hack CG and use your bot to mine bitcoins
Zenoscave: lol JFB
eulerscheZahl: shh, not in public chat
Zenoscave: Was the chat server down for a while?
eulerscheZahl: yes
Zenoscave: I had a could not connect to server for a good 8 hours
eulerscheZahl: the old certificate issue is back
Zenoscave: It was nice
eulerscheZahl: to force you to write more code
Zenoscave: It helped
struct: inneficient in what way?
eulerscheZahl: too bad we have discord now
struct: performance?
Zenoscave: yes struct
JFB: In 2 h my bot (without my actions) went from 79 to 53 in Gold. Using sofisticating prediction algoritms (linear regression) I predict that bfore end of the contest my bot will be on 1 position in Legend
struct: I can tell you that my code that averaged aroudn 1mill-900k sims on STC average is now doing 500k with same code
numb3r5: hey, whats the rank beside people's profile? it says im 12449th
struct: profile is CG rank
numb3r5: based on experience?
eulerscheZahl: based on all contests and multiplayer games
eulerscheZahl: +optim and clash of code, codegolf
eulerscheZahl: but we don't talk about clash of code
numb3r5: why not :no_mouth:
Zenoscave: It's the red-headed step child
wlesavo: Zenoscave oh i figured that one out, the opponent can actually win with one move, but somehow dont want to, so my sim is sure im lost already and goes nuts, not sure it is even important, that shouldnt happen in legend
JFB: We simple NEVER tahlk about Clash of Code !
JFB: ;-)
Zenoscave: How important is endgame prediction in legend?
Astrobytes: The first rule of CoC is never talk about CoC. In World chat.
Zenoscave: Seeing if it's worth sims
eulerscheZahl: very @Zenoscave
tobk: Just noticed another weird thing about my timeout issue. Sometimes the loop that should end after 30 ms will in fact only end after 38 or more ms (according to the print immediately afterwards)
eulerscheZahl: quickly getting some points before it's too late
Zenoscave: Good point
tobk: which is weird, since the loop makes ~5000 iterations, so a single iteration should never take remotely that long (it just tests all the actions which are applicable)
struct: well didnt one legend player reached it with 10k sims?
CommanderCero: hmm my bot always loses because my opponents panic brew in the end... I think i should start simulating them...
tobk: I'm starting to suspect that my timing code might be flawed, using nanoTime In Java (Kotlin, actually)...
Zenoscave: CommanderCero you timeout too
CommanderCero: yeah just saw that
Zenoscave: http://chat.codingame.com/pastebin/b06b7e0d-633c-487f-b4ff-f48ca81d1683
struct: Ok I need to think on eval now
CommanderCero: im 100% sure thats not my fault
struct: Only thing im missing
CommanderCero: 99%
Zenoscave: fuck
JFB: wlesavo - investigation one of my old timeouts I found that I assumed that action_id > 0. And use 0 for REST. So unfortunatelly it was action_id == 0 and instead to learn it - I done REST in the loop :-)
struct: RIP
CommanderCero: nice code
struct: Automaton2020 send help
CommanderCero: xD
Zenoscave: that was a huge mistake
struct: AutomatonNN help a bit too
AutomatonNN: most likely doesnt matter that much
struct: you are probably right
CommanderCero: its cut off
Csipcsirip: endgame detection might be important, but completly ignoring the opponent is still top20 worthy :p
wlesavo: JFB yeah, also checked that, also i though that spell id shouldnt be greater than 100
eulerscheZahl: my rest id is 199 :D
Zenoscave: lol eulerscheZahl
struct: Csipcsirip you got top 20 without opponent interaction?
eulerscheZahl: call 199 as i say to myself
Csipcsirip: yup
struct: ok if i dont get gold i delete
Zenoscave: NEW ACCOUNT TIME
eulerscheZahl: not again
struct: I have 1 day 15 hours to create just the eval function
Zenoscave: if struct doesn't get gold is tobou comming back
eulerscheZahl: i'll give you my old c# but don't delete
Astrobytes: No deleting again please struct
struct: No need euler I just need to find a good eval
eulerscheZahl: you will lose you mod rights
eulerscheZahl: i'll tell thibaud
Astrobytes: And you are a Valued Member of the Moderation Team
tobk: I got to (bottom) gold with just a dumb BFS for best actions...
Csipcsirip: struct if u want a good eval think about the simplest things, that describe the game value well
Astrobytes: "i'll tell thibaud" :D
wlesavo: struct you mispelled legend
struct: wlesavo im not sure about legend
struct: I have a good sim
struct: But im not sure im capable of eval
eulerscheZahl: 1*tier0+2*tier1+...
eulerscheZahl: + game score
Zenoscave: ^ this.
Zenoscave: That should be enough
Washier: yes, think i got that, and only that, right so far
Csipcsirip: ye i didnt want to copy paste my eval but basically ^
tobk: you also take number of learned spells into account?
eulerscheZahl: y
Zenoscave: es
Nerchio: breaking into legend feels pretty hard
eulerscheZahl: :es:
eulerscheZahl: that's the whole point about legend
Nerchio: true
struct: All legends should be like bit runner legend
JFB: It is all due to all submiting good bots befotre start of the Legend
miszu: good morning all :)
Washier: i prepared myself to build a sim this time. To do it right. Didn't work out - how do people built it that fast!?
JFB: They set the bar
ParticleBeam: Probably much practice
miszu: practice and also from beginning they went into sims
eulerscheZahl: this was a simple game egine
miszu: that's what I would do next time.
Washier: hey ParticleBeam. Chestbump
ParticleBeam: Hello
struct: this game is simple, yet I dont know how many times i rewrote
JFB: struct - maybe you rewrote due to siplity of the game?
miszu: my difficulty is to compute all possible moves and apply the move to your state. Lots of corner cases I find
Washier: i did have a root canal on monday, but no excuses. i will do better next time
Washier: but have to add, its a good game. you can get into silver with if-else and clever sorting
ddreams: keep trying different approaches after the contest ends too
kovi: +1 for -1 rest
eulerscheZahl: ?
kovi: rest action code
Washier: <grabs popcorn>
eulerscheZahl: ah, the ID?
miszu: wait that's a good question. How you represent the rest as an action?
Washier: <chewing>
struct: represent?
struct: how I apply it?
miszu: ok so
Csipcsirip: cds.reset()
miszu: I have a vector of Action that represents all legal moves. But it doesn't include the rest.
ParticleBeam: Depends on your structure I suppose. I just create an action with a type I call "REST"
Zenoscave: same ParticleBeam
miszu: yeah it's a hacky way I thought when I went to bed
ddreams: I created an action with id 255, type REST
struct: Action(ActionType::Rest)
eulerscheZahl: class Rest : public Spell
ddreams: or 254 for REST and 255 for WAIT (to avoid crashing where I don't have any legal actions)
miszu: why would you use wait anyways
miszu: rest is just wait but bettre
ParticleBeam: REST is always allowed even though it gives an error message
struct: https://i.imgur.com/UmjYK2f.png
struct: (╯°□°）╯︵ ┻━┻
ddreams: makes it easier for me to see when I have a bug
Csipcsirip: nice
Zenoscave: rip struct
miszu: stuck in silver eh
struct: either my eval sucks
struct: or my bfs is wrong
struct: my numbers dont seem right
Zenoscave: bfs?
Zenoscave: what depth
eulerscheZahl: do we need more mods or can we handle it without struct?
struct: Sims: 759761
Depth: 8
struct: no pruning
ParticleBeam: Ive rewritten multiple times and it is all being beaten by a piece of old code I wrote earlier in the week. Its quite depressing.
CommanderCero: try pruning that did the trick for me
Zenoscave: Jesus! Prune and you'll be so good to go
eulerscheZahl: this number makes me jealous
dbf: teaser for raic-2020: https://imgur.com/a/Ml4alHQ :)
struct: and that number is not the full depth
struct: it is that depth + 1
struct: just not fully explored
eulerscheZahl: interesting dbf
Zenoscave: Yeah beam search struct
struct: Shouldnt I create eval before beam search?
dbf: moved to next week, so no clash with codingame
Zenoscave: Well you need Best N moves not first N moves
Zenoscave: so yes
struct: ok, time to take algebra course
eulerscheZahl: oh, CG is the reason for the delay?
Gronahak: @dbf what's raic ? Looks cool !
eulerscheZahl: russian AI cup
ClockSort: struct it's because you aren't excluding duplicate states. i had the same problem. excluding them cuts states by 10x while leaving depth the same or +1
eulerscheZahl: https://russianaicup.ru/
miszu: in c++, how you define a pointer to point to a const variable?
CommanderCero: oh noice follow up for after codingame
eulerscheZahl: similar to codingame, they have a contest once per year
struct: yeah, i think I can also exclude duplicate states
dbf: eulerscheZahl, no they wanted to start today but have some issues with test system, so moved for next week
struct: I think is something ill ad to the list
Gronahak: ty
ParticleBeam: const int* var?
ParticleBeam: As apposed to int* const var which is a constat pointer to an int
miszu: that's what I did but still it complains
Washier: why have a pointer to a constant var?
CommanderCero: you can move the const around
CommanderCero: means the same
ClockSort: better to have a pointer to a pointer to a const var
miszu: easier to read
dbf: there should be more details onTuesday, so I'll post something here :)
miszu: I'll put a snipet 2 sec
miszu: http://chat.codingame.com/pastebin/f338fecc-4ee7-43a4-87a2-2a379c57ac9c
Gronahak: @eulerscheZahl thank you for what you said yesterday in the chat I gained ~200 ranks because of it :D
ParticleBeam: https://stackoverflow.com/questions/21476869/constant-pointer-vs-pointer-to-constant
ParticleBeam: Apparently
CommanderCero: is players an array of pointers
miszu: players is an array of player
miszu: size 2
miszu: me and the enemy
CommanderCero: try changing the * to &
eulerscheZahl: i have no idea what i said 5 minutes ago :D
Washier: or lose the const's
miszu: I can do this: const Player p = players[playerPos];
miszu: but it does a copy
Washier: weisbier?
CommanderCero: yeah or const Player& p
miszu: I want just a ref
miszu: ohh
CommanderCero: problem is if you access an array you do not get a pointer
Washier: ^^
Nerchio: euler do you use tests in your bots
CommanderCero: if you want an pointer you have to do const Player* = &players[index]
struct: I removed vector and my performance :rocket:
miszu: ohh okay
miszu: I forgot that pointers need a storage
CommanderCero: thats why i prefer references Player& p = players[index]
CommanderCero: thats always possible, but references cant be overwritten with other references
miszu: const Player &p = players[playerPos];
Washier: never liked references. but i'm a m$ monke
miszu: this doesn't do copy constructor?
CommanderCero: nope
miszu: okay ty
CommanderCero: that is equivilant to pointers
CommanderCero: except it cant be null
Washier: references confure people
miszu: eggsellent
miszu: I gave up on trying to reach gold but I want to improve my c++ knowledge
miszu: and learn some optimizations
Q12: I'm so badly want to be in gold :tired_face: I am in rank 108, I am sooo close!!! :weary:
Zenoscave: Submit's slowing
ddreams: references are much easier conceptually than pointers
miszu: ddreams true
ddreams: think of it as an alias for the same object
Washier: dunno, i like pointers. but i'm no pro
miszu: I remember in java how teacher says: you pass by value or by refence. Taht's a lie. All is by value
ScifiSpirit: but aren't some of the values references? :P
ScifiSpirit: if not all of them (i don't know Java)
ScifiSpirit: so all are values, which are actually references
miszu: thing is, if java is doing by reference then you could make a swap function
ddreams: different concept for reference, so that's confusing
miszu: true
ddreams: cleaner in C++
Washier: my backgroun is in in moving data quickly. cant afford to copy stuff
miszu: once you understand a bit of C++, I enjoy it more than java
Gronahak: @Q12 me too man... I've peaked at rank 6 in silver I'm tweaking some magic numbers hoping to make the cut :D
miszu: java is like noob firendly. They simplify the complexity a lot by not allowing a lot
Washier: no mem management yes
miszu: yeah that's one of the thing
ddreams: no operator overloading, or did they add that in the end?
miszu: no you can't
Zenoscave: I don't think so ddreams
miszu: at least in java 8
Zenoscave: maybe in java 200
miszu: at work we still use java 8
ddreams: absolutely hated writing mathematical code in Java for that reason
Q12: @Gronahak
I cross fingers for you, I know we can make it!!! :grinning:
Washier: come to the dark side. C# allows almost everything
ddreams: C# is Java done right
Zenoscave: or C
Washier: even pointers
miszu: or assembly
miszu: no rules
Zenoscave: Why not both?
miszu: no such thing as const, private
Zenoscave: por que no los dos?
eulerscheZahl: Nerchio 06:55PM
euler do you use tests in your bots
as in: unit tests?
eulerscheZahl: no comment :(
Washier: I like C. worked on weather station data loggers before in C - so clean
Zenoscave: https://www.codingame.com/share-replay/508190408
wlesavo: i only put random asserts here and there
Zenoscave: oops with the null move
Nerchio: guess not then euler :D
miszu: next c++ question
miszu: my function returns something expensive
miszu: but I don't want it to make a copy at the end
Washier: pointer
miszu: somehow to use the local variable
CommanderCero: c++ optimizes that, return temporary values wont be copied
CommanderCero: note temporary
Csipcsirip: or just use global variables :p
CommanderCero: yeah but then you have 1 global variable for each function, although c++ optimized that years ago xDF
Washier: make it simple, that function is repsonsible for creating the object, and returning its adress, calling code to use and destroy
miszu: I thought the local variables get copied to a not local variable and return it
Nerchio: i tried C++ and the only way i figured out was to move unique ptr or use shared pointers to avoid things going out of scope when i need them lol
ddreams: global variable doesn't need to reserve new memory every time though
CommanderCero: https://stackoverflow.com/questions/12952295/c-returning-temporary-objects-confusion
CommanderCero: ddreams thats a good argument
miszu: CommanderCero yeah that was my understanding
miszu: there is no hacky way to return a temporary and avoiding undefined behavior?
ScifiSpirit: yep, return value optimization might trigger there. Google it to see how it works. If it doesn't apply in your case, you could either pass a reference to object you created outside of the function, or use shared_ptr or something
CommanderCero: argh i sent you an post form 2012 my bad
CommanderCero: sorry i was distracted
miszu: I don't want to create backdoor
CommanderCero: so since c++ 11 there is the move operator
ScifiSpirit: or maybe you could also move construct another object
CommanderCero: meaning if your class can be moved
CommanderCero: it wont be copied when returning by value
miszu: while ago I read something if you use const in the signature method you could return the reference of temporary and still be allowed
Washier: why so complicated peeps?
CommanderCero: all objects have move constructor by default
miszu: ohh
miszu: okay
CommanderCero: you can also quickly test it by printing in the copy constructor
miszu: but I return a vector
miszu: still works?
CommanderCero: yes
CommanderCero: it will move the pointer to the array
CommanderCero: nothing more
wlesavo: smth wrong with submits right?
CommanderCero: found it https://stackoverflow.com/questions/17473753/c11-return-value-optimization-or-move
emh: wlesavo I'm watching my submission. submitted a while ago
emh: 27 minutes ago
Zenoscave: 34 min ago
Zenoscave: at 50%
Washier: busy busy
miszu: apparently returning by value is fast.
miszu: so no need to worry
wlesavo: i mean for legend it really looks broken, you cant move from the bottom
The_Auditor: the event page has a nice airtable with "useful resources". is there a was to save or download that ?
wlesavo: i would understand if it would be just me
ZarthaxX: emh what did you do????
ZarthaxX: wtf
eulerscheZahl: why ZarthaxX?
eulerscheZahl: submits are getting slow in general again
wlesavo: ok it is just me then
Illedan: 336 results found for «Score - Computing»
ZarthaxX: to get to top silver toad :P
struct: Time to get my spreadsheet
eulerscheZahl: for what?
struct: eval
struct: I dont think I have any bot here with an eval different from -1 and 1 besides CSB
AntiCheap: I can't because I don't have excel
struct: ah csb
struct: stc*
eulerscheZahl: there are other spreadsheets
eulerscheZahl: like Libre Office
struct: ^
miszu: when you use const too much it complains you are not allowed to modify state haha
struct: lol
miszu: I thought const keyword is a genious move.. but now I see it is a torture move
lifetimeLearner007: use const for unchanging constants. For all else, use let
miszu: I like to use const as a guarantee I don't modify something by mistake
miszu: but when I actually need to modify something, it plays trick on me
NinjaDoggy: this is by far my lowest (bot performance)/(time spent) contest :(
lifetimeLearner007: it gives error only for re-assignment, not for monkey patching. just don't reassign
NinjaDoggy: coding in c++ wasn't worth it...
NinjaDoggy: takes me like 30 minutes to find an index out of bounds error :(
ClockSort: ha
miszu: NinjaDoggy, you would need it anyways
ScifiSpirit: :D
miszu: next time, start c++ right away
NinjaDoggy: -_- simply not true
ClockSort: in C# you can spend 0.1ms on that bounds check instead :)
geppoz: submitted 42min ago, 75%
Washier: ^^
CommanderCero: std::vector has the at() function for bounds checks
NinjaDoggy: next time it's java until last 2 days, C++ rewrite at the end
miszu: oh right. [] operator does undefined behavior if out of bound
Washier: not managed
CommanderCero: yeah quite annoying so I use at in the beginning to prevent that
miszu: I was doing a practice here and for 10 minutes I was like... why it's not printing cerr!!???? but then I found out I didn't access [] properly
miszu: # the more you learn
ClockSort: are cg servers running out of memory? I'm timing out on move 1 while mallocing
dbf: ClockSort, try to do malloc after reading of first input line
NinjaDoggy: debugging experience: http://chat.codingame.com/pastebin/4b030cbb-7945-476b-89c9-ddc47ab78bd9
pmor: Make sure you're not allocating too much. I found that out last night when my bot kept timing out for no reason.
ScifiSpirit: lol, yeah NinjaDoggy
NinjaDoggy: do you calculate how much memory you use?
NinjaDoggy: 768 mbs is a lot
AntiCheap: Fuck I think what I found is bayes
NinjaDoggy: but not infinite
miszu: first instnct would be race condition but here you can't
pmor: I don't, though I do allocate some large pools
miszu: at work, I had similar issue where the error was random. Turns out it wasn't caching data properly and you do some randomness in code so it doesn't recompute like previously
ClockSort: if you exceed memory, it times out, or gives error?
AntiCheap: Could bayes be related with your "eval" function?
pmor: I was getting time outs. Timeouts went away when decreased the size of my pool
miszu: why using pools here?
pmor: avoid allocation in bottleneck code
ClockSort: got it. yes, that was the issue. Too much allocation.
NinjaDoggy: do you not calculate your memory usage? XD
kovi: 26?
https://www.codingame.com/share-replay/508221358
kovi: first i thought it was crash/timeout
AntiCheap: Is bayes formula related to statistics?
eulerscheZahl: i saw a 26 match earlier the week. still very rare
eulerscheZahl: especially 26 with that high score
AntiCheap: He could've drown path till win probably
miszu: how the code in ide gets saved?
struct: press play my code
struct: or submit
miszu: ok thanks
miszu: I am coding the apply a learn tomes
miszu: so many things to handle at once
eulerscheZahl: there is no submit button anymore
Zenoscave: TEST IN ARENA
eulerscheZahl: i'm testing https://prnt.sc/vnhuuq
struct: only 12 submits :(
Csipcsirip: 621 submits here
Zenoscave: 197
struct: Save the servers
Zenoscave: Had +1K on OoC
Zenoscave: that was a long month
Zenoscave: about 1 every 40 minutes on average
struct: lol
struct: I dont even have that many for CvZ
struct: and i spammed a lot
eulerscheZahl: and I felt bad already :D
struct: Why so many submits though?
miszu: I checked my UTTT cote, I finished 5th in silver
eulerscheZahl: because of growing frustration and the hope that anything randomly works
Q12: Is "play my code" button doesn't work?
struct: you even have to solve captchas
Csipcsirip: and actually sometimes it works
blasterpoard: kovi thx for letting me know; at one point, my search shows that it can win at turn 25, but it didn't choose that variation
Zenoscave: Struct. Nothing else to do during quarantine
Zenoscave: and was in a statistics class
Zenoscave: so I used submits for a project
blasterpoard: also, in that 26 replay, I got 5 spells I rate very highly (among top 10). and they also have some synergies between them
struct: sample too small, should have submited 10k times
Zenoscave: blasterpoard did you offline spell sorting?
blasterpoard: Zenoscave yes
Zenoscave: struct... every ~4-5 minutes. for a month strait
Zenoscave: I'd be banned
blasterpoard: that's what got me to legend; last nght I ran a simulation and woke up to 20GB of data to parse
struct: 20 GB?
Zenoscave: What's the worst spell? Out of curiosity
Illedan: Id=19
Zenoscave: (0,2,0,0) ?
blasterpoard: I don't remember the worst one, but the second worst one was 1,1,0,0
struct: 0, 2, -1, 0?
Illedan: Not sure, I only have the ID on mine + the score
Zenoscave: ... is there not a 0,2,0,0
blasterpoard: the format in which I have it in my bot is not exactly readable by a human
struct: yes id 15
kovi: i planned to run simu tonight
kovi: but before i check top for another closet...
Zenoscave: closet is gold still i think
Zenoscave: 1st gold
blasterpoard: pairs of spells that have the best effect on my bot are
blasterpoard: http://chat.codingame.com/pastebin/8ef1a9bc-3f34-483a-90b0-d381b058ed4b
kovi: yeah, he is the locm winner known for his excellent draws
aCat: ^^'
wlesavo: dam and im still picking 5 first spells
eulerscheZahl: my worst is BBB => Y
eulerscheZahl: no idea which ID that has
struct: 37
JFB: wlesavo - so you now know how to improve - download replay of blasterpoard, do rank on these replays, implement the rank:-)
struct: oh no
struct: it s0
struct: id 0
eulerscheZahl: and then OO => BBGGG
struct: 5
blasterpoard: I think OO=>BBGGG was my worst
blasterpoard: not sure though
eulerscheZahl: they are very close in my list
eulerscheZahl: best is BBG
blasterpoard: I disagree on that quite a bit
eulerscheZahl: then that's your opinion exclusively :P
blasterpoard: -2, 0, -1, 2 is the best
Zenoscave: I agree with blasterpoard
Zenoscave: now it's not exclusive
NinjaDoggy: lol
blasterpoard: you just need to make sure you have something to start with
eulerscheZahl: maybe i should revise how often my bot rests
eulerscheZahl: that can affect the numbers
eulerscheZahl: but that one is pretty high on my list too
eulerscheZahl: still below BB=>GG
blasterpoard: that's 2nd place for me
E_pur_si_muove: while walking through the search space, don't calculate with rest, just remember that you should had rested here
AntiSquid: i have a formula for mine, overoptimized there instead of elsewhere
NinjaDoggy: how are you guys even precomputing these...
NinjaDoggy: 42 choose 7 is like 2.7 mil
Zenoscave: why chose 7?
NinjaDoggy: depends on how many turns you learn initially
NinjaDoggy: but 7 is a good number i think?
NinjaDoggy: maybe 5 or 6 is enough
NinjaDoggy: 4 learns seem low though
NinjaDoggy: seems*
JFB: I do not know how they calculated but I think that even very simple aproach will work:
JFB: do a lot of runs
NinjaDoggy: yea I'm saying how to store results on CG
NinjaDoggy: maybe that's why they only ranked each one
NinjaDoggy: and pairs
NinjaDoggy: and not all combinations
JFB: I think so
Zenoscave: Yeah just individ or pairs I think
JFB: Only 1 or maybe pairs
NinjaDoggy: ok...
Zenoscave: For Learning Choice
NinjaDoggy: I'm going to try for combinations of reasonable storage
NinjaDoggy: 42 choose 3 seems ok...
tobk: Man, those timeouts are really getting on my nerves! They seem to increase, and are never reproducible or in any way understandable. Could this be caused by the server? are you experiencing similar?
tobk: Or is it just the Java/Kotlin get-time functions being super-imprecise? I read something like that about currentTimeMillis, but nanoTime should be good, right?
JFB: After chabge in my bot I can set time = 42 ms and receive no timeouts
JFB: C++
Nerchio: tobk how many object do you create in a round
AntiSquid: if rand * 10 % rand() * 2 == rand() * 5 timeout the player
AntiSquid: you get the idea
Zenoscave: crashes. Can't multiply method with int
AntiSquid: lol
tobk: Nerchio not sure, but my loop usually does ~5000 iterations before the timeout, so a single iteration (before the timeout is checked again) should not that that long
tobk: you think it's garbage collecting at the end of the search-function?
Nerchio: yeah but when GC decides it wants to clean your garbage you can't say no :D
tobk: good point
Nerchio: it's possible that objects from multiple rounds accumulate i one garbage collection
Beresta: I just forcefully GC up to gen 1 every turn
Beresta: to aviod taht
Nerchio: i don't think you can force GC to do anything, at least here
AntiSquid: this is best contest to have a prune array actually
Zenoscave: prune array?
Beresta: in C# it seems to be easy, GC.Collect(1, GCCollectionMode.Forced);
no idea about java tho
Nerchio: there is System.gc() in java but does it do anything? nobody knows
Beresta: heh I see :D
dbf: it should, I used it at step 0 to avoid timeouts on next steps
Zenoscave: Only if the hamster in the wheel stops running
dbf: but later I disabled object pool and don't need it
Beresta: then, well, you could try to avoid producing new objects at all (reuse old ones) :D
Nerchio: yea i create 400 000 nodes in turn 1 and i dont have problems
Nerchio: but kind of limits stuff you can do anyway
Nerchio: if you want to create new lists and stuff
Beresta: yea, annoying
dbf: cool, I tried to preallocate 200k nodes and had timeout on step 2-5
wlesavo: have 2mil also no timeouts
dbf: :(
Nerchio: its probably something else dbf
Nerchio: if the nodes never get cleared out
kovi: closet in legend, gratz
Beresta: well not sure how java gc works, but if its generational - first 1-2 collections could take a while, till most of the objects migrate to later generations and won't be checked every gc
dbf: Nerchio yep, but I tried several times with the same result - not sure why, but with object pool gc is triggered on some step where I have only 50 ms and there is a timeout
Nerchio: that's interesting Beresta actually, cause my bot takes time to ramp up and only in later rounds it can process a lot of nodes in my search
Nerchio: so maybe thats the problem too
AntiSquid: this submit looks too good, where's the mandatory loss streak Automaton2000 ?
Automaton2000: i don't need to know how to do it in a new window
dbf: it would be cool if CG adjust GC settings so gc is not triggered for java apps when there is still a lot of free memory
Beresta: you can't adjust it from running app at all?
ClosetAI: Thanks @kovi , it was a bit harder this time :)
dbf: no, only cmdline params
Beresta: eh, unfortunate
dbf: yes, I allocated only 200k objects, have 64 megs/768 megs used and there is a gc already
Nerchio: well since alocating all the objects turn 1 and using bitset in my search i stopped all timeouts and my turn finishes like 0.3ms after i want it
Nerchio: but its a lot of steps to not timeout :D
struct: poor java users :(
lifetimeLearner007: I'm stuck in bronze. Any suggestions how to approach?
casmith789: bfs should get you out of bronze, or a good heuristic approach
Gabbek: Correct me if I'm wrong (as that's usually the case): search is only half of the job, you need a decent state evaluation score since otherwise you'll end up with empty hand after crafting each potion? I guess I shouldn't try to do anything when I'm sick since I just dropped to the end of silver trying to "improve" my crap :(
dbf: lifetimeLearner007, what is your approach now?
struct: yes, search is nothing without a good eval
tobk: Hm, GCing not at the beginning of each turn (while waiting for inputs/opp's turn), but still occasoinally timeouting
lifetimeLearner007: I'm randomly taking spells to cast and if the new ingredients are closer to any potion, I reward it.
ClockSort: you can't GC during opponent's turn. your program gets no clock cycles.
lifetimeLearner007: but its doesn't target a specific potion to maximize the points though
tobk: just now, in the very first turn, the first line of my debug output was printed (showing my loop took exactly 900ms (first turn), but then the second line (with the actual result) is not printed
lifetimeLearner007: how is bfs applicable here?
tobk: I use BFS to find a path to any potion (and possibly beyond) instead of searching for paths to potions one after another
Rhyme_McAddames: hi Im new to this but can we use python on here?
dbf: lifetimeLearner007, start from your current state and produce child nodes for each possible action (cast, brew etc) use bfs to find node with good score
struct: yes
Rhyme_McAddames: Ok good thank you @struct
lifetimeLearner007: @dbf, child nodes mean updated states after applying action?
tobk: BTW, few days ago when my code timeouted there was a message that I did not read all the inputs, now that message does not come any more. does that mean my code timeouts "differently" now, it is there another reason?
tobk: *it is -> or is
struct: it showed that message when you timeout
struct: Happened to me aswell
dbf: lifetimeLearner007 yes
tobk: So it's a server-thing that that message is not shown anymore?
struct: It was due to the timeout
struct: So yeah, related to server
Rhyme_McAddames: does anyone use python?
lifetimeLearner007: This is what I'm doing currently. Can someone point out my mistake or suggest to improve?
lifetimeLearner007: http://chat.codingame.com/pastebin/163f0463-86e0-496a-b900-93919ed40d56
tobk: Rhyme_McAddames sure, just have a look at the leaderboards (are you talking about the contest or in general?)
Rhyme_McAddames: in general
dbf: lifetimeLearner007 you do some random search of valid sequence. try to build ALL possible sequences and use bfs to find best one
AntiCheap: Use bfs for each potion or brute force
emh: what's your scoring function guys? do you use absolute score, average score, some combination or lots of ifs maybe?
emh: average = absolute total price of brews / turns to make
BrunoFelthes: 13/41 legend players submiting
sadakatsu: I'm not using a scoring function. I use distance to potions plus a simple comparison rule between goals.
emh: ohh hmm ok
Beresta: sadakatsu what you describe is exactly scoring function :D
AntiCheap: I used "missing items" as initial score
AntiCheap: Worked better than expected
tobk: hm, I might have solved my timeouting by just removing my not-very-extensive System.errs. Next goal: make bot suck less.
yhyoxx: goodluck
lifetimeLearner007: yeah, removing std errors output cleared my timeout issue too
BoBot: haha, nice topdeck by boss
https://www.codingame.com/replay/508281755
game have it the exact spell needed
BoBot: to not lose*
NinjaDoggy: unlucky
NinjaDoggy: outskilled by topdeck
emh: 19 wins in a row yayy, maybe I'm finally going to Gold
geppoz: BoBot 8 initial LEARN ? wow
BoBot: it is a recent change, might not keep it
BoBot: and yea, the limit is hardcoded :/
mojojojo: what was the reason to switch to C++ from C# @eulerscheZahl
ClockSort: speed, i'm sure
SirLoxley: no more GC
mojojojo: ahh ok
DToTheE: Are a lot of the strategies people are employing around optimizing their memory-usage and speed?
DToTheE: There are similar games that try to build up combos of moves, might be a smart approach.
particleflux: yay, when pprof show 10% of time used in the actual logic, the rest is memory allocations and garbage collector :S
DToTheE: Thattts grosss lol.
DToTheE: I'm using Python 3, so I suffer a lot from things like that.
DToTheE: I used a lot of objects, and still do, since they help me think through things, and most interviews they expect you to have clean code with nice classes.
particleflux: maybe I'm just too stupid to read that chart
particleflux: Using go tho
DToTheE: Go has structures, they're lighter if I understand correctly.
particleflux: Yes, I thought so
DToTheE: Well, I guess I can try using a big ol' array of integers to represent game states, but thats not ideal..
particleflux: yeah, ugly to work with
particleflux: I wonder.. can we use SSE/AVX instructions?
emh: particleflux yes you can
emh: at least in C++
struct: yes its possible
particleflux: Would probably be helpful for these delta calulations
Though in C++, I guess the compiler would do so automatically with -O3
emh: if you make the loops simple enough for the compiler to understand yes
emh: might have to specify march=sse2 or something as well
emh: I have pragmas when I do it
emh: on CG you have to use pragmas ofc
struct: yeah, I dont think O3 alone will do it, but im not sure
struct: At least on CG
geppoz: DToTheE I dropped python when i started to have 30ms spent on the instruction: newboards=[]
geppoz: :D
DToTheE: Yes! I have something like that.
DToTheE: Where it is resetting the game states its examining.
particleflux: -march=native would do all that stuff normally.. depends a bit on how new the CPUs are tho
geppoz: one tip: do it just after sending output, instead of do it at start of your turn
emh: #pragma GCC target("avx2")
struct: #pragma GCC option("arch=native","tune=native","no-zeroupper")
particleflux: Might have to recode things in C++, with a dumb bfs I'm only getting ~350 nodes in 40ms with go :S
particleflux: Or maybe try hard-coded arrays instead of adjustable slices
ClockSort: that's the key, particleflux, allocating a lot of new memory inside your loop will be slow(ish) in any language
ClockSort: but you will also get better performance from C++
particleflux: Well either way I can't get rid of these allocations completely if not pre-allocating a huge chunk
SirLoxley: you can try to reuse objects
particleflux: doing that for the easy parts, with go's capacity vs size for slices
Not gonna easily solve the part of allocating storage for new nodes itself
particleflux: Well, it might, with a pre-allocated nodes array
particleflux: That's kinda like implementing your own memory manager though :grimacing:
SirLoxley: this competition clearly puts any implementation relying on garbage collector on a disadvantage
SirLoxley: well ... take it as a non-functional requirement ... you are dealing with a hard real time problem
SirLoxley: missing the deadline = fatal
Nerchio: :joy:
DToTheE: Yeah :P I didn't realize that. I'm so used to walking into interviews, and your algorithm has to be show a nice use of objects, and have a good time complexity, even if its not necessarily blazing fast.
Counterbalance: hence the bloatware we have today
therealbeef: how can any business make money with something that is not time-critical?
Panagis: does "int actionCount = in.nextInt();"
phord: Chrome crashed and lost my changes. :angry:
Panagis: has excecution time 20ms???
phord: What are the gcc or clang switches needed to compile locally?
Panagis: does int actionCount = in.nextInt();
has execution time 20ms from the 50ms
Counterbalance: phord -std=c++17 and I recommend -Wall
therealbeef: @panagis does that include the waiting time for something to appear on stdin?
Panagis: yes
phord: Got that. What about libs?
phord: (Looking for a shortcut)
therealbeef: i assume that waiting time is not part of the time budget
Counterbalance: see codingame.com/faq for the rest.. not needed locally
phord: Thanks.
therealbeef: @Counterbalance when i tried to use c++17 features, it doesnt compile on CG
Counterbalance: it should.. they've been using it for a while now
Panagis: http://chat.codingame.com/pastebin/6f3ef4fd-06e9-40ac-a341-4e8142447884
Panagis: Time=133
Panagis: omg
struct: ofc
struct: you must start the clock after reading first input
Counterbalance: Panagis you're timing the other bot and the referee
SirLoxley: bloatware right there :p
Panagis: @counterBalance thanks
Counterbalance: np :) It's something that probably should go on the faq
Counterbalance: therealbeef i'm curious - what c++17 feature isn't working?
therealbeef: not sure anymore, tried something a few days ago
Nemesis386: hi
therealbeef: @counterbalance oops, seems I was wrong it supports 17 indeed
Counterbalance: a few years ago they switched from 14 to 17
CDA-20285-Gabriel-Misiurny: Hi ! I'm trying to contribute but the stub generator displays: http://chat.codingame.com/pastebin/cf503682-dc28-44b5-a143-c42b664fd7a5
Counterbalance: I don't think you're supposed to put a space after the ':'
CDA-20285-Gabriel-Misiurny: indeed, it worked for N
CDA-20285-Gabriel-Misiurny: now i have
"InvalidType: {"variable":"D","type":"double","line":2}"
Counterbalance: it's supposed to be float.. see the reference: https://github.com/CodinGame/codingame-game-engine/blob/1af6859627b51d42da62a7bdcad93f55a3dad73a/stubGeneratorSyntax.md
CDA-20285-Gabriel-Misiurny: oh gosh ! Thanks !
Counterbalance: not sure if it's the latest one but this was in my browser history :) yw!
CDA-20285-Gabriel-Misiurny: looks fine now !
CDA-20285-Gabriel-Misiurny: :D
particleflux: aaaand gone from 400 nodes to 10k :D
particleflux: stupid thing really, appending to a slice each element separately instead of all at once
particleflux: Should not have made a difference since the memory was supposed to be pre-allocated with go slie *capacity* param anyway, but apparently that doesn't work like I thought
Nerchio: slow submits sad panda
Nerchio: cg doesn't want me to win contest with java?
Poviliukas: it's not only java :(
Nerchio: i know but i am coming for #1 place
Nerchio: and submits are slowing me down
aCat: xD
SirLoxley: First time I beat that silver boss, on a replay I got smashed, though
Astrobytes: woo lazy silver. Do I continue tomorrow or just leave it? :thinking:
NinjaDoggy: do I continue tomorrow or just leave it...
NinjaDoggy: no :)
ddreams: if you leave it, you won't feel bad if you don't achieve anything more :p
ddreams: protect your ego
Astrobytes: I won't feel bad either way tbh ;)
Zenoscave: How defeating that sounds lol
particleflux: continue today? :)
particleflux: still half an hour left
particleflux: :D
ddreams: yeah, I was sarcastic
Scarfield: Nike: Just dont
ddreams: better to try and fail than to not try
NinjaDoggy: hey zeno
NinjaDoggy: how you doing?
karliso: ddreams You are so right
Astrobytes: Not for long, I have the beginnings of a rewrite I don't wanna touch til tomorrow
NinjaDoggy: (dw i'm asking about bot performance, not your well being ;) )
Zenoscave: Server slow down broke my bot. Showed that sims
NinjaDoggy: oof :(
NinjaDoggy: same
Zenoscave: Need to perf optim on eval
Astrobytes: grim server issues this contest
NinjaDoggy: i personally think eval is overrated
NinjaDoggy: at least inventory eval
Zenoscave: No just tome
NinjaDoggy: ahh yea
NinjaDoggy: ok same :(
Nerchio: do you guys have some custom spell evaluation or just normal
NinjaDoggy: planning on doing some heavy sim
NinjaDoggy: for evaluating spells
NinjaDoggy: (offline sim)
Nerchio: i saw euler talking with somebody earlier that some spells are trash basically should maybe avoid them
Zenoscave: Yeah They offlined
struct: As long as I can get gold without that im happy
Nerchio: should share the results with me :smile:
Zenoscave: ok
Zenoscave: after
Zenoscave: ;)
struct: I probably could get gold with 1/100 of the peformance
Astrobytes: No sleep for a day or so then NinjaDoggy ;)
NinjaDoggy: yes there's the spirit :)
struct: Tomorrow is they day of no sleep
Zenoscave: I did that last night
NinjaDoggy: not for me...
Zenoscave: Huge headache now
NinjaDoggy: contest ends at midnight for me
Astrobytes: struct weren't you having like 100's of K's sims
struct: Yesterday I was awake for 22 hours
NinjaDoggy: i don't sleep till midnight anyways...
struct: Astrobytes I have 800k now
struct: with bfs
NinjaDoggy: holy smokes wtf
Zenoscave: Jesus Struct.
Astrobytes: but no results?
NinjaDoggy: explain your shitty algorithm so I can avoid it then ;)
struct: that with very simple eval
NinjaDoggy: 800k? :O
Astrobytes: what about pruning?
struct: no pruning yet
NinjaDoggy: ahh
struct: I shared numbers here before
NinjaDoggy: detecting repeated states is a bitch :(
Astrobytes: Doesn't seem sensible to have that many without pruning
struct: yeah
struct: I need to fix
struct: After eval
struct: is done
Astrobytes: duplicates?
struct: its 500-800k
struct: no duplicates
struct: its just standard bfs
NinjaDoggy: wait wot? :O
NinjaDoggy: how do u check if 2 states are the same?
NinjaDoggy: pls share
struct: no duplicate check*
NinjaDoggy: -_-
Zenoscave: lol
struct: :p
Scarfield: same same xD
Astrobytes: depth?
NinjaDoggy: cuz unordered_map is not it :(
NinjaDoggy: sooo slow
struct: Sims: 827620
Depth: 7
NinjaDoggy: lol
struct: Depth 7 means it started expanding to depth 8
Astrobytes: roll your own NinjaDoggy
NinjaDoggy: yea trying to do that right now
Counterbalance: why are you still silver with that, struct?
Zenoscave: Depth 10 + would be needed to get gold probably. Try pruning
struct: Counterbalance no eval yet
struct: Im working on it
NinjaDoggy: eval isn't a big deal
struct: 10 depth is needed for gold?
NinjaDoggy: opponent simulation is though...
Zenoscave: 10 Beam
struct: I just want gold ninja
struct: not legend
NinjaDoggy: ok idk then...
Astrobytes: If you don't get gold struct I WILL REQUEST that your account be deleted :rofl:
NinjaDoggy: gold is kinda free :P
Gabbek: judging by my crappy results -- eval is really important
NinjaDoggy: you don't even need opp sim...
NinjaDoggy: just literally go for cheapest potion
Gabbek: jumping between 200-2000 changing just eval :D
Gabbek: (in silver)
NinjaDoggy: wait yea try that
JFB: I use bfs - gold with around 400k without pruning (depth suzally 6-7-8)
NinjaDoggy: to go closest potion, tie break by reward
struct: Also my submit does not have this number
NinjaDoggy: hard code first 8 turns to learn
struct: ~this is why its on silver
NinjaDoggy: if you only want gold i think that's easiest way
Bikoko: you ate for 2hours
Bikoko: and didnt sleep :)
Gabbek: NinjaDoggy any advice for state eval? This is what I'm struggling with the most; search seems to do okey
NinjaDoggy: I have super simple eval...
NinjaDoggy: but i'm scared to share cause maybe it's genius :eyes:
NinjaDoggy: jk sure
NinjaDoggy: my eval right now:
ddreams: hmm.. I have "opened 3588 maxDepth 6 0.0410309s "
NinjaDoggy: myScore + score from potions
ddreams: I guess you're not pruning much
struct: no pruning
Zenoscave: score from potions?
struct: This is my first time i use bfs
NinjaDoggy: 1*#B + 2*#G +3*#O+4*#Y
NinjaDoggy: that's it...
NinjaDoggy: yea the potions brewed along the way
Zenoscave: ah right
Nerchio: lots of crashes atm?
NinjaDoggy: oh 1 trick to make it not get stuck at full inventorys:
Zenoscave: Isn't that the same as your score?
NinjaDoggy: yea...
NinjaDoggy: lol
NinjaDoggy: oh i guess i should specify it's current score
NinjaDoggy: oops
Zenoscave: gotcha
Gabbek: NinjaDoggy interesting - I must be doing something wrong then :p
Zenoscave: So what's that trick?
Gabbek: using pretty much the same
NinjaDoggy: don't add the inventory score unless potion score > 0
Zenoscave: https://www.codingame.com/share-replay/508404543
NinjaDoggy: that way it always goes for a potion
NinjaDoggy: down side is... need deep search to use it well
NinjaDoggy: wouldn't work for beam obviously
struct: Astrobytes dont worry i already said gold or delete
Zenoscave: right
Gabbek: doing around depth 15 here with pruning, so I must have some bug somewhere :(
NinjaDoggy: do you account for opponent actions?
NinjaDoggy: like them stealing potions
Gabbek: not at all
Astrobytes: struct if you delete I'm sending a delegation of CG members to Leiria to hunt you down
NinjaDoggy: probably why
Zenoscave: No NinjaDoggy
Gabbek: I ignore opponent atm
NinjaDoggy: deep search sucks without opponent sim
NinjaDoggy: because you'll find some super optimized route
NinjaDoggy: that delays potions
NinjaDoggy: and then ur opp steals them and it's so bad :(
linjoehan: man I'm so lost with this contest atm
NinjaDoggy: I sacrificed like half of my depth
NinjaDoggy: to get decent opp sim
NinjaDoggy: :(
Astrobytes: linjoehan: read the source, it's better than the statement
NinjaDoggy: went from 13-14 no pruning
Zenoscave: I just see what the first possible potion my opp can make and avoid it if I can't make it before or at the same time
NinjaDoggy: to 10 or less no pruning :(
NinjaDoggy: yea I used to use that but...
NinjaDoggy: what if your opp can reach all of them before you?
Nerchio: zenos but that doesn't mean he's going to make the potion and you might miss your optimisation
NinjaDoggy: (usually happens after you brew your potion)
Zenoscave: then just stock up.
NinjaDoggy: hm..
NinjaDoggy: not horrible i guess
linjoehan: lol I know how it works I just don't know how to approach it right now
Zenoscave: It allows for end game swaps
NinjaDoggy: yea it's not bad
NinjaDoggy: but not as good as assuming your opponent will block the optimal one ;)
NinjaDoggy: will optimally block*
Zenoscave: wdym block?
Astrobytes: reflex bot til silver was my goal and it works linjoehan
NinjaDoggy: like go for the potion that blocks you the most
NinjaDoggy: while blocking themselves the least
Zenoscave: Right. Not easy to do that with beam..
NinjaDoggy: debatable
NinjaDoggy: I managed to do it with BFS
NinjaDoggy: just gotta include extra stuff in state
NinjaDoggy: a lot of extra stuff... :(
Astrobytes: wow, Agad e has some lead now. Wonder who delineat e is
NinjaDoggy: recurse smurf ;)
Gabbek: wasn't emil recurse's smurf? :)
Zenoscave: nope
Astrobytes: that was emil no?
struct: D
NinjaDoggy: LOL
Zenoscave: emil was
Zenoscave: a
NinjaDoggy: wait...
NinjaDoggy: wait... it's not a joke D:
struct: emil was re curse yes
NinjaDoggy: -_-
NinjaDoggy: "lemme proto type in rust for fun"
NinjaDoggy: wow
Gabbek: :D
Astrobytes: I wrote both names down last night so I remembered which one disappeared overnight :D
struct: I had no clue that emil was a char from Nier Automata
struct: Same as previous and current re curse avatar
Astrobytes: Could still be trickses, who knows :D
Astrobytes: Oh hahaha, I missed that. I just went by the more obvious clues.
Gabbek: Heya Astrobytes :)
Astrobytes: Hey Gabbek! :)
Gabbek: How're you enjoying the contest?
Astrobytes: Was about to ask you the same lol. Well, I've not been overly motivated tbh, but it's a good game
Astrobytes: You?
Gabbek: I got sick, was lying 3 days in bed - but it didn't help me at all :D was hoping that I would be able to figure out something, but to no avail - guess it's silver time and that's it, hehe. Tried to do some submits today but my search is doing worse than my no-search python bot from day 1
tobk: I guess it's okay, but the last ones were more interesting
Astrobytes: lol Gabbek, sounds like usual CG contest issues :D
tobk: maybe I'm missing some key insight, but not really much of a strategy to consider IMHO. just find the best potion to brew
Astrobytes: tobk: OOC > Pacman tbh
NinjaDoggy: there's minimax strat going on in this game
NinjaDoggy: especially towards end game
Astrobytes: yep
Astrobytes: Is blaster still doing a smitsimax?
NinjaDoggy: doubt it
NinjaDoggy: i don't think smitsi is good for this game
blasterpoard: I am, because I had no time to switch
NinjaDoggy: D:
Astrobytes: Nice job making it work blasterpoard
NinjaDoggy: so... you're telling me
NinjaDoggy: rank 50ish gold -> legend
NinjaDoggy: was just tome picking?
tobk: I am no even considering the enemy at all, and probaly won't start doing so tomorrow. just BFS for best brew path :-D
blasterpoard: yep :)
NinjaDoggy: ok...
NinjaDoggy: i should stop procrastinating tome picking then lol
NinjaDoggy: trying too hard to get by cast/brew optimal
tobk: BTW, what happened to that one guy who was at the top a long time? seems to have vanished off the leaderboard
NinjaDoggy: when it doesn't matter
NinjaDoggy: LOL
Zenoscave: lol
NinjaDoggy: yea emil was banned for cheating
Zenoscave: tobk that was a smurf
NinjaDoggy: unfortunately
tobk: smurf = bot?
Astrobytes: not unfortunately, he's still there ;)
NinjaDoggy: no he got help from recurse
Zenoscave: smurf = alt account
NinjaDoggy: so he got banned
tobk: ah, I see
Gabbek: *confused* :D
Zenoscave: Ninja I just caught on. sorry
blasterpoard: you keep trying to make it optimal, and then all that matters is having spells to reach those last random potions
Zenoscave: Yeah he was banned
Astrobytes: lol
NinjaDoggy: blasterpoard if I guess your tome pick eval correctly, can you confirm it?
Zenoscave: No
Zenoscave: you won't lol
Zenoscave: Or it's just a MC
blasterpoard: oof, you'll have a hard time doing that
NinjaDoggy: ok rip :(
NinjaDoggy: wait is 8 turns of fixed learning actually good?
NinjaDoggy: and the only thing holding me back is WHAT i'm learning?
blasterpoard: I'm learning 6-8 spells, depending on what they are
NinjaDoggy: ok nvm
NinjaDoggy: and you don't do any learns after?
blasterpoard: no
NinjaDoggy: ...
NinjaDoggy: that's so bad...
NinjaDoggy: :P
NinjaDoggy: (the goldie said to the legend player)
Zenoscave: Says the gold
Zenoscave: beat me to it
NinjaDoggy: :D
blasterpoard: yes, my bot is bad
NinjaDoggy: honestly though...
NinjaDoggy: Smitsimax might have potential
blasterpoard: it doesn't
NinjaDoggy: depending on your "modification"
NinjaDoggy: 100% sure?
Bikoko: how to calculate score of clash of code in leadervoard bruhs?
blasterpoard: not 100%, but it's close
Bikoko: average of score of every clash?
Zenoscave: TrueSkill rank Bikoko
Zenoscave: An adaptation of elo
Bikoko: uh i see thanks
sadakatsu: Well, it's more than that. My favorite part of TrueSkill is that it determines individual contributions to team performances, so you can use it for every kind of competition.
Zenoscave: I'm aware lol. Just easier to compare it to elo
Nerchio: ok be back when servers are alive(hopefully this century)
Bikoko: what should i do t improve my COC score?
Zenoscave: win
particleflux: lol
particleflux: easy as that
particleflux: :)
NinjaDoggy: this is factually correct
NinjaDoggy: winning on CG is proven to improve score
NinjaDoggy: something i'm failing to do atm :(
Bikoko: play more game or get more score in every game?
therealbeef: have more fun
Astrobytes: I think a smitsi type approach might work later in the game, but by then it's almost just a minimax
geppoz: lol, when i pick a lost game from arena and send to IDE, I win
Astrobytes: oh, chat scroll sorry
Nerchio: servers are bad and there is more timeouts or its my bot?
Astrobytes: Servers slow yeah
Astrobytes: Probably related
geppoz: and why when i send to ide, i didnt timeout?
geppoz: ide on different servers?
Nerchio: tbh just ignore timeouts at the moment
Nerchio: servers are wonky
geppoz: im trying to get gold, when i send lost matches to IDE to improve, there I win, so hard to ignore... ;)
Whiskee: what if I just send WAIT and let everyone else time out first
Whiskee: new strat
geppoz: :D
struct: You will win maybe 5% of the games
Whiskee: well that would be a 5% improvement
Whiskee: today's been terrible, every little change made things worse
theycallmedavid: geppoz maybe the opponent has somthing random in their code which sometimes work better for them and sometimes not
geppoz: it is not only one... i see 4-5 lost in battle list, all lost for my timeout, and when i send to ide, i didnt timeout... :(
Nerchio: my bot is timing out as well im pretty sure it's servers fault atm
ddreams: does it also time out if it's out of memory?
struct: yes
ddreams: thought so
ClockSort: anyone else having timeouts?
Zaphus: and for other exceptions like going outside the bounds of an arrray, etc
Nerchio: ClockSort massive timeouts yeah
NinjaDoggy: um... I wrote my own hashmap and am kind of confused... :(
NinjaDoggy: the first few turns it's really slow but then it starts getting faster and faster
NinjaDoggy: until it's about 2x faster than unordered_map...
Beresta: why would you write your own :O
NinjaDoggy: it's C++ not java, hot spotting isn't a thing right?
NinjaDoggy: because I explore 3x more nodes without a hashmap detecting collisions
Counterbalance: maybe it's new/malloc for resizing in the beginning?
NinjaDoggy: ?
NinjaDoggy: no my version is slow and speeds up
Counterbalance: does your hashmap allocate memory when you add items?
NinjaDoggy: no
NinjaDoggy: it's fixed memory
Nerchio: you switched from java right?
NinjaDoggy: yea using C++ now
Nerchio: cause java has warm up stuff but in C++ idk
NinjaDoggy: yea exactly..
Nerchio: i think the answer to everything in C++ is to use pragms
Nerchio: pragmas
NinjaDoggy: i have them slapped in
Nerchio: at least that's what the chat is saying :D
Beresta: are you sure in your measurements then? :)
NinjaDoggy: so... idk
NinjaDoggy: yea i'm just counting how many times I add a node to queue
Astrobytes: constructors?
NinjaDoggy: is that... a thing?
NinjaDoggy: i'm pre allocating everything I think?
Counterbalance: maybe it's just that your map/cache gets filled more and more, and thus detecting more and more duplicates?
NinjaDoggy: i thought maybe that's why but
NinjaDoggy: i reset everyturn
NinjaDoggy: and i make sure there's no collisions
NinjaDoggy: so collisions take longer(to find a new index)
Astrobytes: Yeah but are you setting things in the constructor tho, it can be a hindrance on CG if you're allocation shitloads of stuff
Astrobytes: *allocating
NinjaDoggy: i don't mind sharing the part i'm doing:
NinjaDoggy: http://chat.codingame.com/pastebin/0fb212e4-4d74-43f2-9ae7-a15d38ab8183
NinjaDoggy: that's my hashmap
NinjaDoggy: it should all be initialized before the program starts no?
NinjaDoggy: or like before the function gets called
NinjaDoggy: I mean... I'm down to use the first turn to warm up the function, but I would like to know why it needs a warmup...
Beresta: very likely just because of cpu cache
NinjaDoggy: https://www.codingame.com/share-replay/508444754
Beresta: if you hit same chunks over and over again
NinjaDoggy: here's a replay of it happening
NinjaDoggy: N = number of unique nodes
NinjaDoggy: E = total edges from the nodes
NinjaDoggy: C = call count
NinjaDoggy: first turn had tragic amounts(< 10k calls)
NinjaDoggy: but towards the end, it warmed up to > 100k
Counterbalance: if it's global, then it's probably allocated in the .bss section, which is page-mapped by the os to the elf image. Which in laymans terms means that the memory gets activated when it's used..
NinjaDoggy: ok so if I make a function that loops through all my memory
NinjaDoggy: and call that turn 1
NinjaDoggy: It'll be good?
Counterbalance: I think so.. it's about 190Mb by the looks of it..
particleflux: or malloc it in turn 1 if this is the issue
NinjaDoggy: yea, I calculated the size assuming i give it roughly 200mbs
Counterbalance: if you init that before you get any input (so before turn 1) then you should get a speed increase in turn 1
NinjaDoggy: yea i don't care about turn 1 XD
NinjaDoggy: Learn tomeIndex 0 ;)
ddreams: I really doubt that's the problem
NinjaDoggy: (for now)
Astrobytes: your're not allocating in first turn?
NinjaDoggy: i have it just global
NinjaDoggy: is that the issue?
Counterbalance: yeah
NinjaDoggy: so i need to malloc those?
Counterbalance: hmm.. maybe.
NinjaDoggy: but then I can't use the nice
NinjaDoggy: []
NinjaDoggy: :(
Astrobytes: nodes = static_cast<Node*>(aligned_alloc(64, sizeof(Node) * NODE_MAX));
Astrobytes: over-complex example but you get the idea
ddreams: I mean, just memset it all to 0 if you want to touch it
NinjaDoggy: ok and that would work too?
NinjaDoggy: and this is in general I assume...
ddreams: I doubt any of this will make a difference
Counterbalance: me too
NinjaDoggy: then why does it have this "warm up" affect? :(
ddreams: something totally unrelated to that it's a global
NinjaDoggy: hm...
ddreams: perhaps the statistic calculation is wrong
NinjaDoggy: it's not
NinjaDoggy: i change from unordered_map<long long, Node*>
NinjaDoggy: to this
NinjaDoggy: no other change
ddreams: perhaps your inserts are expensive
NinjaDoggy: it's not
NinjaDoggy: and it wouldn't matter i don't think?
Cappefra: what happened to emil?
NinjaDoggy: if this is the only thing i change
NinjaDoggy: emil got caught cheating so he was banned
NinjaDoggy: got help from reCurse
NinjaDoggy: ;)
Cappefra: ouch
NinjaDoggy: lol it's reCurse smurf XD
ddreams: if your inserts are expensive, it will appear faster and faster since more of your time will be spent looking up stuff
NinjaDoggy: sorry not sure I follow
struct: This blue and green ingredients must be a pain for colorblind
theycallmedavid: Who is reCurse?
NinjaDoggy: but looping through all my arrays at the start
NinjaDoggy: seemed to have solved the issue
NinjaDoggy: :/
NinjaDoggy: I guess i'll do that with my other global arrays too...
NinjaDoggy: wtf
Astrobytes: to initialise them?
NinjaDoggy: yea
struct: theycallmedavid CSB rank 1
struct: STC rank 1
struct: WW rank 1
NinjaDoggy: took around 70ms to init my hashtable
struct: FB rank 1
NinjaDoggy: CoIF rank 1
struct: list goes on
pb4: FB rank 1 ?
pb4: Damn
Astrobytes: I try to init the nodes when I have to NinjaDoggy
struct: sorry pb4
pb4: When ?
Zenoscave: lol hi pb4
pb4: Hi :)
theycallmedavid: struct thanks, I don't know what are these though
Zenoscave: Competitions on this site
Zenoscave: Coders Strike Back == CSB
struct: 05-09-2020 pb4
pb4: ah damnit
pb4: With a large margin
struct: yeah
struct: Do you think he managed to apply NN there?
NinjaDoggy: what's FB stand for sry?
struct: fantastic bits
NinjaDoggy: ahh
pb4: ^
NinjaDoggy: i'm so bad with acros :(
struct: best multi on website imo
NinjaDoggy: even though i kjnow all the multis and contests
pb4: struct : very probably
Zenoscave: blasterpoard Can you confirm a single Item positio on your list?
blasterpoard: kinda
Zenoscave: (3, 0, 1, -1) or Id 7 i think
Zenoscave: around middle?
Zenoscave: Maybe slightly above
blasterpoard: let me check, getting that info is not that easy
Zenoscave: ok thanks
Zenoscave: Just wondering if I'm on right track
blasterpoard: id 7 is near the bottom
blasterpoard: on range from 0.0 to 1.0, somewhere around 0.2
Zenoscave: Ok thanks
Scarfield: do you get a better bot with isolated valuation of spells? doesnt it depend too much on your other spells
blasterpoard: but I don't evaluate them independently of each other
Zenoscave: Oh that makes more sense
blasterpoard: that's why I needed to quickly write a script to tell how good it is based on my data
Scarfield: was wondering, makes more sense that way yea :)
NinjaDoggy: wait holy crap...
NinjaDoggy: I've been shooting myself in the foot
Zenoscave: how so ninja
ddreams: you still have feet?
NinjaDoggy: looping through all my global arrays on turn 1...
ddreams: braggart
Scarfield: I really like this contest, sad i havent had more time to put into it
NinjaDoggy: gives me like... 10x speedup
NinjaDoggy: wtffff
ddreams: that's insane
NinjaDoggy: ??????????
Zenoscave: Branch predictions great huh
Zenoscave: Yay warmups
NinjaDoggy: i've been using C++ wrong....
NinjaDoggy: no wonder the speed ups weren't that massive
ddreams: I guess Counterbalance was right
NinjaDoggy: wait so...
NinjaDoggy: if I want to do something like
NinjaDoggy: int arr[1000]
NinjaDoggy: is there a way to do that?
NinjaDoggy: and keep the []
NinjaDoggy: instead of making it
NinjaDoggy: int* arr
NinjaDoggy: and then turn 1 malloc?
Zenoscave: why malloc?
Zenoscave: just use global
NinjaDoggy: apparently global shits on performance...
NinjaDoggy: for example?:
NinjaDoggy: http://chat.codingame.com/pastebin/877598c6-c13d-4b32-8c1e-37d8dd9b7444
Counterbalance: even with int * arr you can still use []
NinjaDoggy: without the warm up
NinjaDoggy: my code runs 10x slower on the first turn
NinjaDoggy: yea I know but... i want to see
int arr[SIZE]
instead of int *
NinjaDoggy: cuz... java :(
NinjaDoggy: pointers look scary
egaetan: why 2688 ?
NinjaDoggy: don't worry about it :)
Zenoscave: Max that can fit?
NinjaDoggy: no it's specific
NinjaDoggy: it's an exact number
NinjaDoggy: will share in post mortem i guess XD
egaetan: NinjaDoggy you missed the 4 spells always here ?
NinjaDoggy: ?
NinjaDoggy: what do you mean?
NinjaDoggy: it's basically impossible to guess why the number is what it is
egaetan: why you don't have 2946944 ?
Astrobytes: you don't need to warmup C++ btw
NinjaDoggy: because spells doesn't ahve anything to do with it
NinjaDoggy: :)
egaetan: NinjaDoggy ok, i take it :)
Zenoscave: 2^7 * 3 * 7^2 * 11 * 13
NinjaDoggy: don't need to warm up C++ but apparently declaring stuff globally doesn't actually initialize the memory
NinjaDoggy: on the right path XD
egaetan: i found easier to do C from java than c++
NinjaDoggy: wrong numbers though
NinjaDoggy: :O
NinjaDoggy: faster too? by a bit?
Zenoscave: 2**7 and 7 **2 no?
NinjaDoggy: nah
Zenoscave: ah 1001 = 7 * 11 * 13
struct: /donate cpu
Zenoscave: You know that arrays in c++ are contiguous?
NinjaDoggy: yeh
NinjaDoggy: so do I just need to do
JohnCM: wow my c++ can get to depth 10
Zenoscave: int[10][10][10] is same size and locatio as int[1000]
NinjaDoggy: array[size-1] ?
JohnCM: but the issue is not the depth now
NinjaDoggy: oh so it's pointless to do my own math?
Zenoscave: yup
NinjaDoggy: D:
Astrobytes: yes
NinjaDoggy: but I can reuse the result right?
NinjaDoggy: like for my 3 arrays
NinjaDoggy: i can compute it once
Zenoscave: int[128][21[1001] wold work fine
NinjaDoggy: and use it for all 3
Astrobytes: yes
NinjaDoggy: opposed to [][][][] 10 times
Astrobytes: wow yes
NinjaDoggy: and similarly, if I want to access the position+1
NinjaDoggy: i can add 1 instead of [][][][p+1]
NinjaDoggy: :P
egaetan: int |64][42][1001]
Zenoscave: Ah fair
Astrobytes: lol
NinjaDoggy: ok time to spill the beans
NinjaDoggy: [1001][32][6][7][2]
Zenoscave: it's your bloom filter
Zenoscave: no?
Nerchio: what is this abomination
NinjaDoggy: it used to be beautiful...
NinjaDoggy: when it was just [1001][32] ...
NinjaDoggy: and then other crap went into the state :(
NinjaDoggy: you'll see in post mortem :(
Astrobytes: oh bloom filter, nice
Zenoscave: I look forward to it
NinjaDoggy: i was calculating optimal paths up to depth 40
NinjaDoggy: in under a second
Zenoscave: whoa
Zenoscave: with all 82 possible actions?
Zenoscave: or pruned?
NinjaDoggy: all cast/brews
Zenoscave: hm
NinjaDoggy: pruned but same result as no prune
NinjaDoggy: Counterbalance, if I wanted to do the same thing with structs
NinjaDoggy: would I need to loop through each thing in the struct
NinjaDoggy: or just one of them, and that would force the memory to be allocated
Counterbalance: I usually just use memset
NinjaDoggy: o
NinjaDoggy: ok i'll do that instead ty
Counterbalance: #include <string.h>
memset(foobar, 0, sizeof(foobar));
NinjaDoggy: and this works regardless what type of array it is
Zenoscave: just make sure the foobar isn't a pointer
Counterbalance: yeah :)
Zenoscave: sizeof (int *) != sizeof(int[10])
NinjaDoggy: um...
Zenoscave: stupid emoji
NinjaDoggy: lol
Zenoscave: int * )
Astrobytes: <cstring> you heathen :P
Counterbalance: struct { .. } foobar[SIZE]; for instance
NinjaDoggy: wait if my node array
NinjaDoggy: my node struct*
NinjaDoggy: has an array of node pointers
NinjaDoggy: how does that work?
Counterbalance: they'll all just be treated as bytes and set to 0. So it'll be an array of nullptr.
Zenoscave: a pointer is an address so 4bytes always on a 64bit arch
Zenoscave: I misunderstood I think
NinjaDoggy: ok cool
NinjaDoggy: so the struct would calculate its size properly
NinjaDoggy: and everything gets zeroed
ddreams: if you have an array of pointers, only the pointers would be zeroed
NinjaDoggy: yea i know
Counterbalance: It might screw up your Node queue[..] though; that is initialized by the compiler already, calling the default constructor for each items
NinjaDoggy: that's what I want anyways
ddreams: so it has nothing to do with "o the struct would calculate its size properly"
Zenoscave: 8Bytes****
ddreams: if you have an array of structs, sizeof(arr) would depend on the size of the struct
NinjaDoggy: no i mean
NinjaDoggy: i have an array of pointers
NinjaDoggy: inside my struct
Zenoscave: https://cdecl.org/
Astrobytes: lol
ddreams: right, so yep, everything inside it would be zeroed
ddreams: doesn't matter what it contains
Astrobytes: alright I am exhausted for today, catch you on the flipside everyone. Good luck :)
Counterbalance: It should already be zeroed though, if you don't specify any initializers
NinjaDoggy: c ya Astrobytes
Counterbalance: nn astro
NinjaDoggy: yea but, isn't int arr[1000] the same?
NinjaDoggy: it should also be zero'ed if declared globally right
Counterbalance: yeah
NinjaDoggy: and it wasn't actually initialized/zeroed correctly(i assume)
NinjaDoggy: for my hashmap
ddreams: according to Counterbalance, it was zeroed, but not loaded into memory
NinjaDoggy: right
NinjaDoggy: that's the word
NinjaDoggy: and i'm trying to make sure my nodes are also all loaded into memory
Counterbalance: you could check that in your for loop.. it should all be 0 already. I think writing to that memory just causes the kernel to map it to actual memory all at once in the beginning instead of generating a bunch of page faults later
karliso: you can do that more efficiently
NinjaDoggy: :O ?
karliso: I have used "mlockall(MCL_CURRENT);"
karliso: I think it was for similar purpose or something here.
particleflux: hu, interesting
ddreams: hm, perhaps also add MCL_FUTURE then
NinjaDoggy: what do I need to include for it?
ddreams: <sys/mman.h> perhaps
karliso: I think I have
karliso: #include <sys/mman.h>
- include <unistd.h>
- include <poll.h>
NinjaDoggy: linux only?
NinjaDoggy: rip windows users :(
particleflux: windows swaps as it wants either way
ddreams: I put the linux specific stuff within #ifdef
karliso: I am on Windows..
NinjaDoggy: uh...
NinjaDoggy: did I mess up my C++ installation?
ddreams: it's a POSIX call
ddreams: probably not
NinjaDoggy: is it because i'm using mingW?
ddreams: I don't have it in a normal VS2019 install either
karliso: Tbh I am not actually sure if mlockall helped, but I did use it and maybe it did.
ddreams: https://stackoverflow.com/questions/1918785/how-to-lock-pages-in-memory-using-winapi
particleflux: yeh VirtualLock, the basic thing you'll need when hackily writing into other processes' code lol
particleflux: CG servers are probably linux though, I guess?
ddreams: yep
NinjaDoggy: eh... seems like a lot of work to swap between them from CG to local
NinjaDoggy: i'm fine losing 300 ms turn 1
NinjaDoggy: i don't even compute anything turn 1 anyways :/
ddreams: #ifndef TESTING
mlockall(MCL_CURRENT);
- endif
ddreams: define testing when compiling locally
particleflux: yeah that or probably _WIN32
particleflux: my windows days are too far away o.O
struct: damn
struct: have a bug that I cant reproduce offline
struct: ..
karliso: I use #if __GNUC__ and #if _MSC_VER
AllYourMCTS: is rust compiled as release now in ide games?
struct: yes
struct: Finally managed to reproduce :)
particleflux: that was fast
particleflux: :P
NinjaDoggy: ok well.. submit time!
NinjaDoggy: i think I just 10x my bot performance...
NinjaDoggy: by writing own hashtable and forcing memory
NinjaDoggy: (lower rank incoming)
struct: ok c++ is crazy
Counterbalance: random unexplainable timeouts incoming
struct: how can 1 eval break my bot?
struct: Counterbalance is it possible to explain it?
struct: Depth: 255 only if
struct: :D
NinjaDoggy: https://www.codingame.com/share-replay/508504953
NinjaDoggy: wait a sec...
NinjaDoggy: :(
Nerchio: :joy:
NinjaDoggy: it's timing out on turn 1 trying to memset everything...
NinjaDoggy: :( :( :(
NinjaDoggy: wait karliso come back... help with the lock thing :(
ddreams: why not go back to your basic array loop?
Counterbalance: if you replay in the ide does it also timeout?
NinjaDoggy: no
NinjaDoggy: it's somewhat rare
Counterbalance: I'm getting a bunch of timeouts too, even though my bot prints the action at 40ms. I guess the servers are acting up.
ddreams: you may still time out if your code is doing cleanup
Counterbalance: it doesn't, in my case. I print the time taken; next line I print the command.
struct: ok anyone has any idea why i go to depth 255?
JohnCM: haha wonder what it takes to get to gold
struct: What can it be?
struct: access invalid memory?
NinjaDoggy: wait just kidding... timouts aren't rare at all :(
Zaphus: My IDE version self-times to 40ms, and is getting 60ms frames at the moment
NinjaDoggy: probably?
NinjaDoggy: C++ is oof honestly
JohnCM: if you go depth 255, it probably means your condition to break out of your search is not activating
JohnCM: might be a bug somewhere
ddreams: if you time out suddenly when using memset, probably there's something wrong there
struct: ok I think I understand
JohnCM: what does memset do again?
JohnCM: why is it so important
ddreams: sets memory to a value
NinjaDoggy: i didn't submit with loop version
JohnCM: can you memset for STL also?
struct: it goes to depth 255 because, I can only wait
struct: so its realistic :p
NinjaDoggy: lol
ddreams: I've never used memset with STL
struct: What are you trying to do NinjaDoggy?
JohnCM: the bulk of my memory usage is in STL
struct: allocate memory?
JohnCM: unless you all use fixed arrays to store data
struct: or assign values at start?
NinjaDoggy: allocate
struct: but not assign right?
NinjaDoggy: nope
NinjaDoggy: allocate to zeros
struct: this doesnt work?
struct: State* Agent::pool = new State[2000000];
NinjaDoggy: well.. sorta...
phord: don't memset stl structures, though.
NinjaDoggy: it works but
struct: What is wrong?
NinjaDoggy: it's not really allocated
JohnCM: is there significant advantage of memset?
NinjaDoggy: like it doesn't really get allocated until I use it
NinjaDoggy: which slows me down a lot
NinjaDoggy: (previously)
NinjaDoggy: https://stackoverflow.com/questions/42429806/why-the-c-global-variable-not-affect-to-memory-usage-of-program
struct: So you are telling me my code can be faster?
NinjaDoggy: yes...
NinjaDoggy: so i'm trying to force it to all be loaded
NinjaDoggy: by memsetting everything to zero
NinjaDoggy: so it's forcefully used
NinjaDoggy: but sometimes that's timing me out on turn one :joy:
struct: So you are telling me that if I assign default
struct: after allocating
struct: Then it will be faster if i assign it again?
NinjaDoggy: apparently so...
Counterbalance: sched_setscheduler: Operation not permitted :(
NinjaDoggy: that's why i'm going through all my arrays and memsetting them to 0 on turn 1
NinjaDoggy: downside is it limits the amount of memory I can use quite a bit
ddreams: try to use mlockall(MCL_CURRENT);
then
NinjaDoggy: i just tried it on CG
NinjaDoggy: and it compiled and ran
NinjaDoggy: but didn't give the speed up memsetting everything did
NinjaDoggy: so i'm just using slightly less memory now
NinjaDoggy: since i over allocated node array and stuff anyways
struct: (╯°□°）╯︵ ┻━┻
Counterbalance: :poop:
struct: submit was not even close :(
struct: got 15th
Counterbalance: many timeouts?
Michael_Howard: :deciduous_tree:
Michael_Howard: (╯°□°）╯︵ ┻━┻
struct: Counterbalance 0
struct: Maybe my opponents have, I dont
Counterbalance: what was your rank before?
struct: 900 on silver
struct: I feel like my eval is still weak
Counterbalance: and now 15? that's a huge improvement
struct: and I dont prune bfs yet
struct: Yeah, but I want gold, maybe tomorrow I can do it
ddreams: what do you use eval for if you use bfs?
Counterbalance: you're already 13 now.. might get above boss by morning
struct: Maybe
struct: ddreams i use ingredients, rupees and brewed potions
Counterbalance: why are you making 6 stars at the end here, struct: https://www.codingame.com/share-replay/508558738
struct: Best depth 8 brewed 8 score 114
at turn 36
struct: My bot tries to brew 8 potions or even more
struct: He doesnt satisfies at 6
struct: At turn 36 he foud that path :p
Counterbalance: at turn 33+ we have both 5 orders, so all you had to do was brew any potion faster than me and you'd have won.. only 1 star needed
ddreams: why not stop at 6?
Counterbalance: so you're scoring the inventory maybe a bit too high?
struct: Counterbalance the shortest was this one
struct: to brew the nexxt
struct: Best depth 10 brewed 6 score 79
Counterbalance: ah i see
struct: That one is at turn 32
struct: Maybe I need to check the inventory eval you are right
Nerchio: this is going to be one of those games where you get a good search but cant go to legend because top gold is too random? :d
Counterbalance: yeah i thought i spotted something but it looks good.. first brew for your bot is indeed at depth 8
blasterpoard: top gold is fine
struct: Maybe ill set up brutal tester
Skynse: tfw you clash super late in the night and you hear someone walk out of thir room
struct: and check if I can improve eval against myself
struct: ´should be faster
struct: brutaltester works on ubuntu?
Counterbalance: sure
Counterbalance: but i'm getting mixed results
struct: In what way?
Counterbalance: like, 60%-20% winrate locally, but 50 ranks lower in arena
Counterbalance: but it does help a bit to see if you haven't screwed up enormously
struct: I see, I doubt it will happen on silver though
struct: but maybe
Beresta: hm, I've put 7 different times that covers 99% of the implementation and it is consistently 12ms lower than global timer on turn
Beresta: could it be print to error stream itself? :(
Beresta: timers*
Beresta: or any other ideas?
struct: What language?
Beresta: C#
Beresta: its not GC for sure
struct: Well CG is having problems
struct: with timeouts
Beresta: nah I don't have timeouts, its +-2ms precisely to my 45ms treshold
ddreams: memory?
struct: ah you mean different timers give different results?
Beresta: yea, sum of inner timers 12ms less than outer timer
Counterbalance: on return from a function, local variables get released. If they are arrays or maps/hashes and stuff that could take some time.
elderlybeginner: I'm disappointed with my contest result. Any tips what it is easy to overlooked and has significant meaning in the contest?
MarekM: debuging without debugger
struct: Do I need maven to run brutal tester? or just to compile it?
ddreams: you run it with java
sadakatsu: el
sadakatsu: ... When did this chat get focus? Sorry about that.
struct: What do you mean?
sadakatsu: I sent the message "el", except I never tried to do that.
struct: must have clicked by mistaske
AllYourMCTS: el
struct: hmm, i changed my bot to stop at 6 brews now
struct: Not sure if I submit or not
struct: oh wait im 10th atm
miszu: question: fast int16 vs fast int 32. Appart from space, would 32 be faster than 16 in certain points?
struct: fastint16 gets the fastest int that is at least that size
miszu: ok so it would be impossible that a fast 32 be faster than faster 16
miszu: fast* 16
struct: depends on cpu I guess
struct: I would recomend against using uintfast
miszu: uintfast no good?
struct: https://stackoverflow.com/a/46960509
miszu: wow
miszu: okay, fast int then
miszu: 16
miszu: I am rewriting my UTTT, I want to reach gold
struct: why fast int?
miszu: fast int guarantees fastest performance
struct: it wont give you any benefit over uint16_t
struct: Why is it faster?
miszu: it takes into consideration the architecture and chose the best int that would be the fastest to do cpu operations
Counterbalance: just tested - it'll be 64 bit on CG (aka linux 64bit)
miszu: it would make sense
miszu: a 32bit would need to be converted into 64
struct: I think using fastint is opening the door to bugs
struct: Just my opinion
ddreams: shouldn't be a problem unless you're targeting multiple platforms
ddreams: probably not going to do much for your rating though
miszu: to get gold in UTTT, bitboarding is important right?
struct: no
struct: for legend yes
miszu: I finished 5th in silver using MCTS
struct: how many sims?
miszu: and thought maybe give a try with simple bitboarding
miszu: very low though
miszu: like 500s
miszu: I don't remember
miszu: it was 8 months ago
miszu: but low
struct: ok, yeah maybe 500 cant climb to gold
struct: But is close enough, dont need much more
miszu: a simple game like UTTT should get you more sims
struct: for legend you need around 20k+- on second turn
struct: rollouts*
struct: not sims
miszu: thus why I want to use this project as c++ exercise
struct: Counterbalance your github repo works on brutaltester?
Counterbalance: yup
struct: Ok I had to check on discord and change one var
struct: I guess its due to my JDK version
Counterbalance: git clone ... referee; cd referee; git checkout cli; mvn package; presto
miszu: is discord chat more popular than this?
struct: "tomato14/11/2020
I fixed it with .map(s -> (Spell) s) before .filter. At least it works now."
Counterbalance: no it's updated, it should work out of the box
struct: I know but I got error
Counterbalance: compilation error?
struct: /home/user/brutal/FallChallenge2020-main/src/main/java/com/codingame/game/Game.java:[530,35] cannot find symbol
[ERROR] symbol: method getId()
struct: I changed a line and it works
struct: and least compiled
struct: dont know if there is a difference between flatmap
struct: and map
Counterbalance: i fixed that in the cli branch, i think you didn't 'git checkout cli'
struct: oh right
Zaphus: did a complete rewrite of my spell learning algorithm, totally different heuristics.
Stabilizes at ~140 in gold
Resubmit of code before rewrite, ~140 gold
I give up :-)
NinjaDoggy: :(
NinjaDoggy: hopefully I'm not about to have the same experience
struct: Lets see if I wake up at gold
struct: .40 from boss
struct: gn everyone
NinjaDoggy: wait struct
NinjaDoggy: ok gn i guess
struct: for what?
NinjaDoggy: if you don't make it tmr
NinjaDoggy: i'm sure with ur # of iterations
Counterbalance: nn struct
NinjaDoggy: it can't be hard...
NinjaDoggy: gn though!
struct: yeah, I was trying to setup brutal tester
Counterbalance: (btw updated my fork; git pull && mvn clean package)
struct: but failed
struct: im not very good on linux
NinjaDoggy: we can talk tmr...
NinjaDoggy: pretty sure there's small things we can do
NinjaDoggy: to massively boost ur performance
Counterbalance: you'll be mid-gold when you wake up struct
struct: yeah, I ont do smart stuff
struct: I just have bfs with mediocre eval at best
struct: What saves me is the number of sims
Counterbalance: > 1 million sims, doesn't save me :(
struct: you have >1mill?
Counterbalance: yeah.. i used to have 6mil with my dfs
NinjaDoggy: wtf...
struct: o.o
NinjaDoggy: can I have some?
NinjaDoggy: just a fraction is good
struct: Do you use anything special?
Counterbalance: i might be counting them differently than you though
struct: 1 sim = 1 move
Counterbalance: yeah
NinjaDoggy: 6 mil?
Counterbalance: plus checking all orders
struct: yeah
struct: how did you get 6 mill
Counterbalance: my sim is basically just 2 asm instructions
struct: ah
mel-hadj: I stuck on silver is minimax a good algo to give a try
NinjaDoggy: it's ok
NinjaDoggy: but hard to make work
struct: minmax might be a bit hard to apply
struct: If I had 6 mill i would try minmax
NinjaDoggy: I "effectively" have minimax
NinjaDoggy: for depth 10ish
Zenoscave: ten brews?
Zenoscave: or ten moves
Counterbalance: the trouble is that the search tree explodes so even 1000mil isn't enough
mel-hadj: what is hard about minimax ?
struct: many states if you dont prune well
Zenoscave: It has simultaneous moves also
Zenoscave: or is that no issue
mel-hadj: so some recommendations !!
mel-hadj: please
The_Duck: I recommend you ignore the opponent. You can get pretty far without thinking about them at all
NinjaDoggy: what about ignoring learns?
NinjaDoggy: is legend possible ignoring that ;)
Counterbalance: nn guys
The_Duck: probably!
NinjaDoggy: nn Counterbalance
Whiskee: I wish writying to the console dor debugging didn't cost cpu time - it's pretty funny that if I stop debugging this round then the pruning works differently and the results are worse
Whiskee: *writing
NinjaDoggy: yea it would be nice :)
guspihl: http://chat.codingame.com/pastebin/8eabf09e-7fcc-4650-b870-1c12166e78f2
