YCS-Venom: how do we know that we are playing with a bot or not?
Severed: arent they all bots? Isnt that what we are doing?
Uljahn: if you're not a bot why do you get captcha?
Perlorodka: @kavandoctor I guess I know why you are asking.. And I simply had to store the demonstration of your amazing skills for future generations. Creating program to convert roman numbers to decimal in 38 seconds, just wow! Too bad you didn't share it afterwards..
GoogleJeff: lol, clearly they are galaxy-brain and type at 1000 wpm
GoogleJeff: it's python so they probably just did "import roman_numeral_clash_of_code_solver"
Velcoro: I'm amazed that some people are actually willing to spend time to exploit an educational site in order to achieve higher rank in some virtual ladder, learning nothing in the process
darkhorse64: What are you referring too ?
Westicles: gotta be clashes
Westicles: or golf
Westicles: Is that why Allis quit?
Westicles: Ah, he's probably referring to a new guy with 58 clashes that is about to pass him.
Westicles: Someone up to shenanigans no doubt
sleepy-coder-101: please join this
Velcoro: I was referring to kavandoctor
Perlorodka: Allis was quite happy with the state of CoC and he was very active, participating in most private clashes.. but then he suddenly disappeared
Westicles: Ah. Well, I mean he is playing against guys who have done every clash at least 20 times. I guess fairplay is a little nebulous all around
darkhorse64: Yeah, when it comes to a point where you copy/paste from your database, it's time to move to something else. Bot programming for instance ?
KiwiTae: Sk0v read the description and hints
KiwiTae: press submit
Uljahn: click Test in Arena
KiwiTae: your not in the ladder u need to submit your code
Uljahn: are there any battles in the Last battles tab on the left panel?
Westicles: maybe a bug, jrke reported same thing on discord
Bitse: I think all battle arenas are currently down. Battles aren't updating for Tron Battle either.
Uljahn: looks like servers've stalled
jrke: yeah battles bug i have reported
Uljahn: understandable, have a great day
[CG]Thibaud: thanks jrke we're looking into it
VizGhar: i though it was only sopra steria challenge :)
Westicles: VizGhar imposteur! pas français!
aetrnm: Hey, guys! What is the shortest programming language available here and how do I actually win "Shortest" game mode?
VizGhar: Westicles :D
jrke: oh new update on CG
jrke: ACTIVITIES AND COMMUNITY nice
Westicles: Still when you click on friends, it doesn't take you to your friends
Satzmann666: Heyo, did anyone else have issues with the ANEO puzzle? Im pretty damn sure my code is correct but tests 7 and 8 fail
darkhorse64: Don't take it personally but for sure, your code is wrong.
jrke: yes there is one minute thing you have to fix to get that perfect in aneo
Passifi: the one with the traffic lights I had the same issue in my case it was something about floating point precision a value was like 59.99999999 or something like that and should have been 60 :)
Passifi: it drove me mad debugging it btw
jrke: yeah minute thing i said
darkhorse64: I do not see any update. Is it available from the home page ?
jrke: "ctrl+shift+r" try it darkhorse64
Satzmann666: thx guys using Decimal did the trick
G-Rom: A/B test for the menu, 50% chance to have it
Kitkat260: any tips for braces
Westicles: Avoid fistfights?
NASIRUDDIN: i am nasir
Kitkat260: nice to meet you i'm kitkat
jacek: sti changed?
Kitkat260: how is everyone today
Westicles: don't you mean hell is?
eulerscheZahl: so, who got the new header already?
eulerscheZahl: i may have. the livestreams have a yellow circle now. rest is the same
BlaiseEbuth: This is visible on both old and new.
eulerscheZahl: what else changed?
BlaiseEbuth: But on the new the categories change
eulerscheZahl: aah, got it in an incognito tab
eulerscheZahl: activites => practice, compete, learn
eulerscheZahl: and community
eulerscheZahl: and discord link in its own community subsection
Astrobytes: Yeah, got it as well.
eulerscheZahl: is it removed from the chat then?
Astrobytes: No, seems to still be there
eulerscheZahl: it's costing valuable screen space :(
Astrobytes: I agree. Should probably mention it.
eulerscheZahl: your job, you are A
eulerscheZahl: i'm B
Westicles: there are probably ads everywhere, we all just block them
eulerscheZahl: i tried to block the livestream with my adblocker
Joseph_Mother: JOE LIKE JOE MAMA
eulerscheZahl: every time a stream ends or a new one starts, it's back
RoboStac: oh, apparently I blocked discord message with ad blocker, was about to say I couldn't see it anymore
eulerscheZahl: my troll sensor tingles
RoboStac: but it's still there on my other pc
Astrobytes: Joseph_Mother: be nice or be kicked.
eulerscheZahl: i guess i should block the discord too :P
eulerscheZahl: perfect, it's gone
BlaiseEbuth: The new header seems not to be displayed in the same way that the old one... It appears on my chat client...
Joseph_Mother: why coc
Joseph_Mother: also why are you telling me to be nice i just helped him correct his words
Joseph_Mother: thats a very nice thing to do
Joseph_Mother: i dont appreciate you profiling me
BlaiseEbuth: I don't appreciate you trolling
Uljahn: what if someone answers negatively but it's actually possible? :thinking:
BlaiseEbuth: What if someone answers positively but it's actuallly impossible ? :thinking:
Uljahn: exactly. why not just try and see for yourself?
Westicles: All I know is Msmits got to 54 in 3 hours playing it manually
Uljahn: MSmits is kinda GA under the hood i guess
BlaiseEbuth: I'd like to say it's possible but with a very good evaluation function, and a non-naive generation of individuals, perhaps helped by some heuristics, and probably with more CPU/Time that allowed on the site...
Kitkat260: thats good
MSmits: I'm more SA under the hood Uljahn, but i did do it manually. I was going to do a beamsearch for it this weekend
MSmits: hope i can finish something that works
jacek: i have the new navigation. and i still can find something else than clashes :(
lopidav: where is the forum at
lopidav: what is happening
BlaiseEbuth: A forum master lopidav has lost...
lopidav: I can access it via history but where is it on the page? Did it move to another site?
jacek: commmunity -> forums
lopidav: oh, that's way better than it was before. Thank you
eulerscheZahl: oh, now i have the new header as well
struct: How much time should I allow per turn for Amazons?
Joseph_Mother: 69 ms
Joseph_Mother: i miss shortest code ones
struct: euler if you block these 2 things the streamer thing will be blocked
struct: www.codingame.com##.container-0-2-47 www.codingame.com##.on.contentContainer-0-2-23
struct: or do you only wanna block the circle that show the number?
Westicles: With ABP is comes back
struct: oh i use ublock origin
BlaiseEbuth: With adguard you can directly select to block on the page
struct: I think most blockers allow that
BlaiseEbuth: I got this : codingame.com##.live-streams-button-container
Westicles: The problem is the three digits change every time you refresh (or maybe when the streams change)
BlaiseEbuth: Ah, this is on the old header.
BlaiseEbuth: On the new header : codingame.com##.live-streams-button-react-container
Joseph_Mother: LUA WHAT THE FUCK
Joseph_Mother: 5.0 == 5
Joseph_Mother: WHAT IS YOUR
struct: that isnt a lua problem
struct: and chill with the caps
Joseph_Mother: Im mad bruh
Joseph_Mother: thats some bs
jacek: oh my
struct: jacek what time do you think I should allow for amazons?
struct: or it doesnt matter much
struct: I get a better answer everytime
jacek: that was my style :(
jacek: how many moves amazon game lasts?
struct: max should be 92
jacek: 100ms is standard
struct: ok, so 2 games it is
struct: I was thinking on making it lower and playing 4 games
jacek: 4 games? what for
struct: I dont know I just thought it could be good
struct: Maybe im wrong
Astrobytes: 2 is enough
jrke: for what?
jacek: i couldnt find nice amazon memes
struct: amazons jrke
JLukeSkywalker: yeah, i cant clash anymore
JeroenB: I have the same issue
JLukeSkywalker: and this is why change is bad
maras: it's laggy as hell today
JLukeSkywalker: dont fix something that isnt broke
JLukeSkywalker: they changed the entire site format
maras: I'm waiting like 5 minutes to run my code
JLukeSkywalker: and now its dying
djmil: yes.. test cases fails on timeout
JLukeSkywalker: try clearing your cache and site cookies so that it loads the new webpage
JLukeSkywalker: helped me a bit
djmil: nope) this is internal site problem
JLukeSkywalker: there is definitely an issue connecting to the server, but the web page issues on your end will just compound it
djmil: i've tryed incognito mode, and it can not open main page :(
djmil: as hell!
struct: yeah, very slow
JLukeSkywalker: last clash i was in, no-one loaded in. 8 ppl got 0%
JLukeSkywalker: good job codingame
struct: Even bigger sites have problems sometimes
struct: dont need to be salty
monkey: Friday deployments ... always fun :wink:
Salted: Maybe it's just me, but the site seems to be down a ton lately. Are they having budget problems?
djmil: it's not only you)
struct: They deployed today, it might have bugs
Oguru: Glad to see i'm not the only one, i was just refactoring my code because i thought it was too inefficient (it probably was anyway)
djmil: Friday deploys.. mmmm... my favorite
NikkiBear: uggggggh LOL
Astrobytes: Oh, not again. lol
eulerscheZahl: the Sopra thing got too competitive, the 95% were intimidated
darkhorse64: The battles have been down for most of the day. It does not help. Saelyos posted a bot 18 mn afer the start and is still #1
Salted: Oh it's getting bad. 503 errors. I'll try again tomorrow.
eulerscheZahl: Saelyos is a strong player
Astrobytes: Yeah I saw the issues this morning darkhorse64. Thought they'd fixed it.
darkhorse64: he is and he is fast
LouieAniez: show saelyos?
eulerscheZahl: and he might had a bot already, at least for the classic multiplayer
eulerscheZahl: even this contest isn't entirely new
Astrobytes: Almost the same as the Klee one
Astrobytes: Or whatever it was called
darkhorse64: I wasn't there
eulerscheZahl: only almost?
eulerscheZahl: I didn't try Klee
Astrobytes: I joined but didn't submit because it was codebusters
eulerscheZahl: wasn't Klee the contest where they leaked all the email addresses?
Astrobytes: I don't remember if it was exactly the same this time around or slightly different.
eulerscheZahl: that's an achievement: host a contest and the only thing I remember is a massive doxxing
Astrobytes: Yeah. Was a real classy move :D
jacek: now lets do the csb league recalc
Astrobytes: the email leak was Klee's fault as well btw
Astrobytes: Not CG
eulerscheZahl: i know
eulerscheZahl: what's that BCC? never used it
Astrobytes: I'm sure I still have the Klee apology email somewhere
eulerscheZahl: i never registered for that contest, so I don't have it
eulerscheZahl: but someone impersonating CG invited me to a paid coding tutoring
eulerscheZahl: and some others too
Astrobytes: I remember that yeah
Astrobytes: I've not had anything like that
eulerscheZahl: probably got my email from github. it's not the one I use on CG
eulerscheZahl: are you on github?
eulerscheZahl: https://github.com/astrobytes this one?
Astrobytes: I've got two, that's mine yeah
Astrobytes: wtf is that
Astrobytes: Oh Thor
eulerscheZahl: i see a thor
eulerscheZahl: yes :D
Astrobytes: I had some short url pointing at it
Salted: Anyone use this before?
eulerscheZahl: i used some offline checker
eulerscheZahl: you download a large file and run a script on it
eulerscheZahl: and with high probability it tells you if you got pwned
Astrobytes: That sounds... interesting
eulerscheZahl: apparently I was but couldn't find my passwords currently in use
eulerscheZahl: wasn't willing to enter my email in some random textbox of a website
Salted: Google passwords seems to keep track of leaked passwords and lets you know if you're using a known password
Astrobytes: Yeah I know right. "Have I been pwned..." - well, you have now :P
Salted: Yeah, I tried my old school one. It's been pwned a lot apparently.
eulerscheZahl: https://www.twitch.tv/soprasteria Neumann speaking in that weird language again
Lanthanum267: i have a problem, when im on https://www.codingame.com/ide/puzzle/temperatures the test cases take like 2 mins each and im often getting 504 error and error #75 "Verification process has elapsed"
jacek: good this chat has prevention to accidently paste your password
jacek: like **********
eulerscheZahl: and he disabled the chat, we can't ping him :(
Astrobytes: Wat. Shocking behaviour!
Salted: So if I paste my password it gets starred?
Salted: Doesn't work
jacek: you can see it but we dont
Salted: Alt + F4 key reveals uses email addresses in tool tips
struct: On amazons should I bother rotating the board when players switch?
struct: Or should just black go first
eulerscheZahl: rotate avatars
jacek: just switch players' positions
struct: But board is symmetric
struct: no need for it :(
eulerscheZahl: do whatever the chess game does
jacek: random openings?
Astrobytes: shuddup you
eulerscheZahl: i downloaded flashpoint 197k flash games waiting for me
eulerscheZahl: and good ones from your memories?
jacek: flash :unamused:
Salted: Is codingame hosted through aws?
eulerscheZahl: CG made contradicting claims
eulerscheZahl: i know that the frontend and email are from AWS
eulerscheZahl: for the games they wanted to switch to Azure
Salted: I'm working on getting the aws developer associate certification. Any opinion on those certifications?
eulerscheZahl: i think it's mostly HR who cares about those
Salted: It's been nice doing it so far. There's a lot of little things I didn't know amazon could do.
Salted: I guess amazon does a lot though
eulerscheZahl: I got a 1 week training 2 years ago
eulerscheZahl: was overwhelming
Salted: It is overwhelming! I'm thinking of switching to coding from engineering.
Salted: I've been an electrical engineer for ~10 years but it's harder and harder to find work that pays anything.
kovi: 1 week training? must be a strong nn after that
eulerscheZahl: seems jacek is influencing multiple users
Salted: Ha! @kovi
Westicles: I thought SW jobs are all in India now. EE should make much more
Salted: When we posted jobs for entry level ee we would get 400+ resumes.
Westicles: I guess it depends a bunch on specialty and location
struct: site slow again
eulerscheZahl: poor Neum4nn
struct: Everything is broken
Salted: Really? My chat was posted to a pastebin?
Astrobytes: Which is also - broken
Salted: haha, oh man
struct: if its long enough it goes to pastebin
Astrobytes: Depending on line breals
Astrobytes: *breaks and length
Westicles: Everything is broken. The stream sounds like gibberish
jacek: isnt this french?
Salted: Well I was trying to say that EE tools have stopped getting updates for about 8 years. No new companies seem be entering the market. I think no new tools means that industry is dying or stagnant. So it's time to bail on electrical engineer.
struct: stream sounds fine on my end
Clement.G.: is there any problem with codingame site ?
Salted: Friday update
Clement.G.: When this will be patch ? :/
Salted: Zero idea
Salted: But its definitely rare to be down like this.
struct: reminds me of contests
struct: The sprint contest of locam was fun
Westicles: perfect time for everybody write a couple more contribs
struct: I cant give legalActions on Amazons
struct: Should I give the number of legal actions?
jacek: why you cant give legalActions?
struct: I mean i cant give them all
struct: for first turn its 2176
struct: Its way too much
Westicles: just give the best one
jacek: alright then give number of actions
jacek: are you gonna give 'random' like chess?
struct: I could
Westicles: So it seems very likely Tryangle Catch gets a 3rd approval today.
jacek: now lets wait till MSmits will give each move % winrate
Westicles: Better to WIP it until the french one is over?
DesmondW: I've always seen it slow on Fridays but not other days. Is that anyone else's experience?
eulerscheZahl: CG servers are looking forward to the weekend too
VizGhar: what is foco?
eulerscheZahl: i'm off for today. good night
Astrobytes: night euler
struct: 25 million games per hour
struct: stop it
jacek: AutomatonNN will you stop him
AutomatonNN: possibility to continue as in a bad code
MSmits: what a stupid waste of calculation time, what's the point of knowing which side has a tiny edge with completely random moves
MSmits: at least do this mcts style, why full random
MSmits: I hope they needed the heating
Astrobytes: Yeah, kinda strange study.
Salted: Probably need a paper
MSmits: you can probably write a paper with less than 1 billion games :P
struct: its only 40 hours
struct: not that much
MSmits: oh right, full random, no search
MSmits: i guess i spent a lot more on my meta mcts :P
MSmits: but at least there is a *point* to that
reCurse: Eye of the beholder
reCurse: The only absolute fact is it's still better than cryptomining
MSmits: guess so
MSmits: hey Astrobytes I presented my project for game design todat
Astrobytes: Oh cool, how'd it go MSmits?
MSmits: started with the yavalath demo and then proceeded to explain my game design
Astrobytes: Yeah Mekole, servers are burning
MSmits: during the talk, i heard through the zoom call: "YES I WON"
MSmits: so apparently they were still playing the demo
Astrobytes: hahaha epic
MSmits: so i asked what level
MSmits: thats basically a 1000 iteration minimax
MSmits: not even depth 2 :)
Astrobytes: no worries Mekole
Astrobytes: MSmits :rofl:
Astrobytes: Hey, they're new to the game though
MSmits: yeah its fine, but its just funny he exclaimed so loudly when its the lowest levek
MSmits: I thought he beat the hard one
Astrobytes: And that was your professor?
MSmits: no, another student
MSmits: professor was being professional
Astrobytes: Ah, you present to the other students too, gotcha.
MSmits: went well overall, i'm glad thats done
Astrobytes: Game design doc went well too then
MSmits: yeah they were impressed i think
MSmits: except the one guy i guess, he was too busy playing yavalath
Astrobytes: Do you have to take questions on it or it's more informal?
MSmits: it's informal but the professor asked questions
struct: Just wait until he comes here and takes the 1st spot on yavalath
Astrobytes: lol struct
MSmits: he did ask about codingame
MSmits: so who knows
Astrobytes: Oh nice
MSmits: my next course is functional programming
MSmits: starts in 2-3 weeks i think
MSmits: elm i think, but it may be different
Astrobytes: A lisp is a lisp is a lisp ()
MSmits: no idea what that means, i still have to learn that
jacek: or haskell
MSmits: not sure, i saw an old version of their course explanation that said elm
MSmits: I havent tried to check on US news for 2 days
MSmits: it's great
Astrobytes: The problems are not going away though.
MSmits: but it's boring enough to ignore now
MSmits: i dont live there after all
Astrobytes: Status quo. Everyone can go back to sleep.
MSmits: thats why he's called sleepy joe
1400179: Did dbd r disclose what he was doing on othello? Just curious
Astrobytes: Yeah, multi probcut
jacek: and 300k weights :?
MSmits: multi probcut ab pruned minimax
1400179: So minimax with linear eval?
MSmits: i guess so
jacek: he also used n-tuples i think
Astrobytes: You sure jacek?
Astrobytes: Was that a subsequent version?
jacek: he was talking about 300k weights. what else could it be
MSmits: the probcut thing uses weights
jacek: but its 2 or 3 params no?
Astrobytes: Yeah, probabilities. You use them to cut...
kovi: but a full random stat can be irrelevant. say if black has a single winning 1st step and all other slightly loosing
MSmits: jacek not sure, i thought it was more
MSmits: maybe multiple lookups get combined to get those 2 or 3 numbers
1400179: He has by far the strongest bot, at least according to my attempts
jacek: we may never know until someone hilights dbdr
MSmits: I knew that reCurse
struct: oldjohn said its the bot that looks more human like
Astrobytes: not that many params but you end up with a lot of probabilities.
jacek: so you beat me reCurse? :(
1400179: Not by much but yeah
MSmits: the reason dbdr is not higher is because he's almost fully deterministic
Astrobytes: Theyr'e used to influence a and b cutoffs, I've forgotten most of the algo
1400179: db dr is the only reason I haven't submitted yet
1400179: My winrates are absolutely abysmal
Astrobytes: It's an excellent bot
Astrobytes: Logistello on steroids
MSmits: sometimes he plays perfectly as far as i can tell, but usually dbdr's bot makes 1 mistake. No more than that
MSmits: when he plays perfectly i draw him
jacek: othello that won 6:0 to human champion in 97
Astrobytes: Yeah. Buro's othello program. Was what he invented MPC for iirc
struct: Didnt oldjohn work with him?
1400179: I'm starting to have a theory why NN eval is not working but it's kinda annoying
MSmits: I'm curious, do share if you want reCurse
1400179: The endgames are extremely easy to compute
1400179: But extremely hard to predict
MSmits: checks out in my experience
1400179: So I suspect all the NN energy is focused on endgames that can be computed in microseconds
1400179: So I need to write an endgame solver
Astrobytes: Yeah that sounds sensible
MSmits: thats not hard to do, but will take you a bit of time
1400179: That's where I notice the biggest deltas of eval anyway
MSmits: it would actually make a lot of your NN's better if you added stuff like that
jacek: i noticed you dont use shortest wins in bt sometimes. meh
1400179: I only bother with what's necessary
MSmits: i noticed you sometimes make a mistake on oware with 7 or so seeds remaining
MSmits: have seen a draw turn into a win fo rme
1400179: It's possible, I suspect oware has the same issue with endgames
1400179: As othello
Astrobytes: Something like that :D
MSmits: yeah makes sense
Astrobytes: Game is bonkers
MSmits: well let me know if you want any help with the endgame stuff, i dont mind sharing some of how i did this with oware. it's a pretty algorithm
1400179: Well I'm doing othello first I guess
MSmits: would you go for a typical ab solver?
MSmits: or a specialized solver
MSmits: specialized would go deeper, but if you dont solve, you dont have a good move
1400179: Not sure what you mean by specialized
MSmits: like proof number search
MSmits: it only proves the win/loss
MSmits: but it doesnt give an eval score
1400179: That's what a solver is
MSmits: but a normal minimax search also solves
MSmits: and if you fail, you still have a good move
1400179: Sure but I'm not talking about search, I already have it
MSmits: i meant with a simple eval instead
1400179: Then I have two problems
Astrobytes: Did you do a proof number solver in the end MSmits?
MSmits: no, but i understand the theory. I was gonna use it for uttt, but never did
1400179: Not sure what you mean by that
Astrobytes: Yeah, it's a bit more work
MSmits: by proof number search?
MSmits: it's a treesearch that prefers the most solving nodes
MSmits: and only solves, but doesnt eval
1400179: Ah i googled proof number solver, serves me right
Astrobytes: I had some papers but I think I've misplaced the links
MSmits: it is usually better than a typical ab or mcts solver search
MSmits: but you won't have a backup move if all you calc time goes into this
1400179: That's the point
MSmits: Angecide no,
MSmits: it's hard to explain over chat, it's a bit weird
MSmits: basically, nodes are just unsolved, until they are solved, no eval or other stats. It just keeps the number of unsolved children
MSmits: or descendants actually
MSmits: let me find a good paper sec
1400179: Intuitively it seems like it shouldn't be better than minimax
MSmits: this one is ok i think
MSmits: comparison with ab on page 21
MSmits: i remember another paper that also compared with mcst solver, but i cant find it now
MSmits: in any case, all PNS variants were usually better than ab or mcts
MSmits: makes sense to me, solving is all they can do, it's specialized for that one purpose.
Astrobytes: yeah there's a couple of papers (probably still by Winans et al) that helped me understand
1400179: Well today is one of those days I'll feel dumber by the end of it
Astrobytes: I feel like that every time I read one of these papers
MSmits: reCurse it's informed ignorance
MSmits: before today you did not know
Astrobytes: hah, nice
1400179: It has to use some kind of heuristic
MSmits: yeah, it counts solved and unsolved descendants
1400179: Does it assume stuff like positions that lost are more likely to lose again or something?
MSmits: with backpropagation
1400179: Ok so then it's heavily dependent on the domain
MSmits: probably, but it has been used in many
MSmits: it might be very effective in othello because othello has the mobility as part of most evals
MSmits: and mobility means more nodes
MSmits: or this might make it anti-effective. Not sure
1400179: I'd rather not tackle more heuristics
MSmits: it's not really a heuristic
MSmits: it's just the only thing that is tracked
MSmits: like mcst tracks wins and visits
MSmits: this tracks solved and unsolved instead
MSmits: so it's quite simple
1400179: I think the tree will cause a massive overhead
1400179: I'm not looking to solve depth 20 positions
MSmits: ohh, my bot does 15-20
MSmits: just from mcts solver
1400179: I want to solve positions that take less time than an eval
MSmits: well that's no problem then if you only need the last 10 or so
MSmits: just use any typical othello eval
MSmits: 10 is real easy
1400179: I want to minimize domain knowledge
MSmits: well it's really minimal, my bot has 7 params and features
MSmits: but ok
MSmits: you can just do mobility only
MSmits: maybe that'll work
MSmits: just avx the thing, there's some stuff you can steal online
MSmits: very efficient move generators
1400179: If I start handcrafting evals then what's the point of going NN
MSmits: reCurse the point is, there is not much handcrafting required since you require so little depth
1400179: So why mobility
Astrobytes: In contrast to other best-first algorithms PN search does not need a domain-dependent heuristic evaluation function to determine the most-promising node to be expanded next . In PN search this node is usually called the most-proving node. PN search selects the most-proving node using two criteria: (1) the shape of the search tree (the branching factor of every internal node) and (2) the values of the leaves.
MSmits: cuz you already have the moves generated. Count them :P
1400179: Sorry I don't get it, use it for what? Move ordering?
MSmits: yeah i guess so
MSmits: or iterative ab?
MSmits: to use with TT
1400179: I think a dumb ab is as much as I can afford
MSmits: why "afford"
MSmits: code size limits?
Astrobytes: This can be done offline
1400179: The point is to let the eval focus on better roi
MSmits: dont you just use the full calc time in those last 10 turns?
1400179: Instead of focusing on stuff that can be solved in microseconds
1400179: But if the ab takes 1ms then it's pointless
1400179: return on investment
MSmits: ohh now i get it
1400179: There is a domain agnostic property, if it takes less time to brute force than eval then just brute force it
MSmits: you want to end the search early
MSmits: and solve
MSmits: i thought you would spend a whole turn doing this
1400179: Well maybe
MSmits: oh that's tricky
MSmits: then you cant do 10 turns
MSmits: too expensive
MSmits: if you want to do it at every leaf 10 turns from the end i mean
1400179: I don't know how fast it can be
MSmits: I wonder if TT is helpful here
MSmits: probably not no... there's some, but not enough
Astrobytes: PN* uses TT
MSmits: it's very helpful in oware, with those low seed boards
MSmits: Astrobytes optionally i think?
Astrobytes: No it's part of its iterative deepening structure
Astrobytes: *multiple iterative deepening
MSmits: reCurse the reason i recognize your end game issue is that when i am hardcoding moves vs old john for example, my bot tends to do the best moves for 10-20 turns in a row sometimes and then messes up right before it can solve, it's quite annoying
Astrobytes: A few other variants use it too
MSmits: so it doesnt need the book until the very end it seems :P
1400179: Well it's also that the game is at its most volatile at the end
MSmits: so there's something volatile about the endgame
1400179: Even though it's very straightforward
1400179: Like imagine if a game is solved at the end by very complex formulas that a computer can bruteforce super easily
1400179: Then why bother learning that formula
MSmits: the best bots i think have 3 separate evals, beginning, mid and endgame
Astrobytes: *evals or approaches
MSmits: there's 1 serious othello bot on gith
1400179: Seems sensible
MSmits: loads of code
MSmits: it has those 3 evals
1400179: I'm not even sure why I'm spending time on this tbh
1400179: but w/e
Astrobytes: I think I've seen that one
MSmits: because you started :P
Astrobytes: Play some Othello
1400179: I don't even like the game
MSmits: tbh it's not my favorite either. I prefer BT over it
1400179: BT is very fun to watch
MSmits: but hey i never played it for real
MSmits: Astrobytes says its fun
Astrobytes: BT is just hardcore pawn...
MSmits: it's just very strategically visual or however you say that
MSmits: no beginner knows whats going on with othello
Astrobytes: It's easier to visualise than Othello for sure
MSmits: same with oware, that seed stuff is very obscure
Astrobytes: If you actually play it (or a variant) physically it's not so hard
Astrobytes: re. Oware
MSmits: yavalath, connect4, BT, these are all games any beginner will be able to "get" in a few minutes
MSmits: not be good, but at least follow
1400179: Maybe I should have done checkers instead
Astrobytes: Oware (and variants) are played by kids around the world
jacek: its a draw-fest
Astrobytes: As well as professionals
MSmits: reCurse go here for checkers: http://www.fierz.ch/checkers.htm
struct: Thats why they play from certain positions on tournaments
MSmits: jacek checkers is mostly drawn because most bots dont know how to finish a game
MSmits: you get infinite repetition
MSmits: even though any reasonable end game db will give the strongest player a win. But it might take 50 rounds and ab bots dont see that far
MSmits: I still have it on my list to put a 3 piece end game db into checkers. I hope that helps a bit
Astrobytes: You can use different variations on ab (or indeed a completely different algo) at various stages in the game remember
MSmits: It just gets a chore to fit all those params, unless you've built for yourself a framework to do this automaticlaly
Astrobytes: And here was me thinking that was the fun part :see_no_evil:
MSmits: well I used to do this with CG bench
MSmits: and sometimes locally depending on the game
MSmits: but I really just need to do it automatically. Get some TD going
Astrobytes: You do have a point.
MSmits: jacek did it and he sucks at math apparently
Astrobytes: No, he hates mathematical notation.
MSmits: o right that was it
MSmits: but thats a hurdle
jacek: i just need to code the math
MSmits: was just saying i need a TD framework
struct: Dont know what to do for UI on amazons
MSmits: what are your choices?
MSmits: and also, what do you mean by UI?
MSmits: it's a bot game
struct: The viewer
MSmits: techically the interface is the input/output structure is why i asked
Astrobytes: that would be the API :P
MSmits: well should it be different from other games in some way?
struct: I guess ill do something simple
MSmits: yeah i guess api is the correct term
MSmits: you can always beef it up while it's in WIP
Astrobytes: Nothing complex needed atm struct
struct: yeah i can
Astrobytes: Yeah, as MSmits said, if something is needed it'll be pointed out
MSmits: I dont know how to even write a search for a game with a branching of 1k :P
MSmits: I need to prune over 90% of that
struct: There are a lot of papers
struct: they say from first 2k moves, 20 are good
MSmits: does it remain that way?
MSmits: 1% being good?
MSmits: hm ok
MSmits: so you hardcode the first move
MSmits: or random from a set of 20
struct: end game is hard
Astrobytes: There's a reason they use it for research ;)
MSmits: with a NN, you have just enough time to lookup each possible move :P
MSmits: without even a tree
Astrobytes: jambokampy: sorry, it's a known issue, devs are aware
MSmits: jambokampy it's laggiest mode
MSmits: gonna go sleep. Hopefully tomorrow i can get some cotr working
MSmits: see if i can beat manual solutions
Astrobytes: lol, masochist. gn MSmits.
Astrobytes: hey zhoubou, how's it going
Astrobytes: Always time for CG... well, not really, but you know
Astrobytes: Yeah, not much to report from here tbh
Astrobytes: When is your work exam?
Astrobytes: Plenty time then
Astrobytes: I got it :)
Astrobytes: Enough to work some CG into the schedule outside of other life things at least ;)
zhoubou: I'm wondering... How do you handle learning programming? Or anything for that matter. With so many things to go into, how to not get caught up in a rabbit hole with so much information available, so many different courses and tutorials. So many programs and learning paths. Omg lol
Astrobytes: Well, when I decided to properly learn I went from the ground up.
Astrobytes: Theory first.
Astrobytes: The internet was a little less helpful than it is today. There's a lot of helpful stuff out there but also a lot of absolute crap.
Astrobytes: Which parts of mathematics are you lacking in?
Astrobytes: And what's your goal with programming?
struct: for some stuff you dont really need math
Astrobytes: ^ It depends what you want to achieve and what level of knowledge you want
struct: if you want to make a NN you probably will need math
zhoubou: Re programming goal, I am thinking about it career-wise in the long run. In the short run, I want to build my portfolio, so some projects. In even shorter run, I'm here on CG, trying to improve.
Astrobytes: For general CS stuff it's graph theory, discrete maths, linear algebra, maybe number theory, that kind of thing
therealbeef: don't forget that programming is a practical discipline; so just program a lot, and search for answers to specific questions when you encounter them
Astrobytes: ^ yep
struct: Everytime I try to read the papers that are posted here
struct: I cant understand the math
Astrobytes: Many, many programmers are not overly well versed in aspects of CS, hopefully they know the domain they are working in :)
therealbeef: You can also learn a lot from having your code reviewed by others
Astrobytes: Weren't you doing a CS degree struct?
Astrobytes: Personally, I always recommend everyone doing mathematics to some level or another.
struct: But I still struggle with it
Astrobytes: lost in a vector field :)
struct: I have phases where my brain doesnt work at all
struct: dont know why
jacek: or struct field
zhoubou: I am also interested in math, and would love to explore it, but I don't know to which extent, since I'm also interested in a ton of other stuff and I'm not sure how much time I can spend on math alone.
Astrobytes: struct, I've been there
Astrobytes: Still happens tbh
Astrobytes: You have to 'chunk' things into what you *need* to learn in order to learn what you *want* to learn. And spend some time thinking about that beforehand.
Astrobytes: Break "what you *want* to learn" into things you *need* to learn
Astrobytes: Pick a goal. I'm not saying you need to think "I'm gonna make a self-driving car!" - start tiny!
Astrobytes: Small projects with clear goals, try to figure what you need to learn to accomplish that.
Astrobytes: Yeah, CG puzzles and multis are good for that in a way
Astrobytes: But a lot of stuff here you might never even use in a dev job
Astrobytes: True to an extent.
Astrobytes: No, not at all
Astrobytes: Great place to practice languages and learn algorithms.
Astrobytes: It's just likely you won't be using a/b minimax or MCTS in your day-job :)
therealbeef: CG is also great to experiment
therealbeef: and to learn optimization
Astrobytes: Yeah, the time constraints really make you push yourself
Astrobytes: But definitely learn elsewhere too. CG is more 'practice-based'
therealbeef: you won't learn software engineering (= professional) here ;) just programming (=hobby)
Astrobytes: Yeah 100%. You can learn a lot here, an awful lot - believe me - but not how to be a dev.
Astrobytes: "paid spaghetti" :D
Astrobytes: Sure zhoubou. Let your portfolio and knowledge do the speaking, not your learning sources.
therealbeef: true, ZtimE
Astrobytes: Yes. Delve into projects, delve into the language(s) and go for it.
therealbeef: Most new hires I get in my team are pretty rough, but that doesn't matter. we give them 100+ comment in their first code review and the polishing starts :D
Astrobytes: Failures are just lessons.
struct: Just code like you are playing coc and you will be fine
Astrobytes: struct :rofl:
Astrobytes: Yeah, don't say *that* in an interview :P
Astrobytes: Even little apps/tools that help you in your current position
Astrobytes: Right. So build on that.
Astrobytes: Keep the motivation up :)
Astrobytes: Eh, get that out the way first!
Astrobytes: Also, I meant to ask, how is the rebuilding effort going in the earthquake-affected areas zhoubou? (Petrinja iirc?)
Astrobytes: I heard there were some tremors recently as well, just to help everything
Astrobytes: Still, it stops work being done. Paperwork eh. Bureaucracy > lives remember! :/
Astrobytes: Hey, you need a job.
Astrobytes: Can't fix the state sitting in a field.
Astrobytes: I get the same feelings working in pharma regulatory. At least it's not the big companies.
Astrobytes: And before anyone asks I've got nothing to do with SARS-COV-2 vaccines.
Astrobytes: I just validate or invalidate things.
Astrobytes: I don't work for anyone btw, they just hire me on a consultancy basis.
Astrobytes: Sounds really fancy and well-paying but it really isn't.
Astrobytes: Especially since Brexit.
Astrobytes: Before my health issues (my back) - and the whole pandemic issue - I was assistant manager at my local conservation centre, which was awesome.
Astrobytes: What perks? Sitting in production rooms analysing labels? :D
Astrobytes: It's not the most glamorous job. Checking samples, validating new cold-storage facilities, staring really hard at tablets looking for faults, writing SOPs
Astrobytes: Ah, Standard Operating Procedures
therealbeef: Did you finish those TPS reports?
Astrobytes: lol, Office Space right
therealbeef: I'll send you the memo again
struct: I suck with colors, everytime I change something a new illusion appears
VizGhar: Yeah guys... All hard puzzles done :metal:
Astrobytes: struct: on Amazons?!
Astrobytes: VizGhar: hah! Well done sir :)
Astrobytes: Maybe you just need a break struct!
Astrobytes: Aren't you just doing it on a chessboard struct?
Astrobytes: What's the issue?
struct: I dont know what colors to use
struct: Those colors make the border look thinner
struct: On some parts
Astrobytes: It's hard to say if that's an illusion or if you're off by a pixel
Astrobytes: Why not drop the lines?
Astrobytes: lol, nah. No worries struct, just think simple first.
struct: not sure if site is still slow
Astrobytes: Seems a little better?
Astrobytes: Anyway, out for tonight, gn and cya tomorrow
struct: Thats CG for work
struct: no CG
Jasperr: Yeah, that happens every day for ~15 minutes at UTC 02:42
Jasperr: UTC 02:15*
Salted: 100 person clashes?
Salted: This is a bit mad
struct: you managed to join one with 100 people?
Salted: It had slots for 100 people and listed it as 5/100 people
Salted: Luckily not 100 people joined before it started. That would be madness!
struct: Max I joined was on ~80
struct: It crashed shortly after
struct: you couldnt submit at all
Salted: Ha! I guess the site is still a bit buggy after the update
struct: Nah, these was long ago, back when that feature was released
jjhiggz: hello everybody