struct: DomiKo I dont really see any bug on that replay
mzbear: i'm so fed up with MCTS after Blocking that i'm writing UTTT bot super casually lol ... just heuristics to see which moves lead to immediate guaranteed loss/win
mzbear: turn timer says 0.025ms per turn on average ... so refreshing :D
mzbear: it would be funny to have alternative bot boards with ridiculously low time limit ... like 5ms per turn
struct: mzbear the minimum time sdk allows is 50ms
struct: per turn
struct: Also some languages would struggle quite a bit with 5ms
mzbear: i suppose the platform can't reliably handle managing games that fast, either. any lag would ruin it
mzbear: oh no, there's no way i can defeat silver UTTT boss with simple heuristics ... do i really have to make a search function :cry:
struct: Hard to do well on it without a search
mzbear: i kinda wanna reach gold without a search. just slapping MCTS on this doesn't really feel like a worthy challenge
struct: Well you can try minimax
struct: but it performs worse
mzbear: that's also a type of search
struct: a didnt read 1st part properly
struct: Well you can try teccles heuristic
mzbear: or maybe i should say ... i dont want to build a forward going move tree
struct: Teccles heuristic is to force opponent
struct: to play on the same board you just played
struct: Some people use it on the start of the game
Westicles: Isn't UTTT the ultimate "fastest mcts wins" contest?
mzbear: hmmm... well, that's an interesting way to get some moves on the board, but it isn't exactly intuitive whether it's a good idea in the long run. could be exploitable
mzbear: westicles, i refuse to believe that!
struct: Quite a few people on legend use it
mzbear: i'm sure it would perform quite well, yes
Westicles: Bitboards required
mzbear: i'm already doing that
mzbear: i even got some useless BMI2 usage going on :D
struct: There is quite a few stuff you can do with MCTS
struct: Im sure its just not fastest wins
mzbear: mcts would require quite a bit of tuning, there's no way i'm doing that by hand in the web editor ever again
mzbear: my heuristic bot still loses to silver boss every time, but at least it manages to fight back a little. average move time 0.03ms
mzbear: but that made it play weaker against everyone else in silver. heh. i'll need to think a bit before writing code
mzbear: oh wow, i had typoed a hex constant in one of my BMI2 optimizations, fixed it and suddenly my bot plays much better
mzbear: it wasn't capable of reading the board correctly ... which boggles the mind, because it still got into silver league
jacek: bmi2 in uttt? w00t?
mzbear: it's due to the board representation i'm using
jacek: also, teccless heuristic is good opening. i added it and managed from top 20 to top 10. at least its good for player 1.
mzbear: for the global board, instead of separate occupation maps, i use 0 for open, 1 for full/blocked, and then 2/3 for player states. so i can test against 0 to see the subboard is still open
mzbear: which necessiates _pext_u32(mask,0x15555)&_pext_u32(mask,0x2AAAA) if i want one player's pieces extracted from that
jacek: :be
jacek: :thumbsup:
jacek: do you have other heuristics?
LightFury01: hey guys
mzbear: this thing is ONLY heuristics :D
RaunakCodes2625: HI
RaunakCodes2625: WHICH LANGUAGE
jacek: oh
RaunakCodes2625: I AM IN PYTHON
RaunakCodes2625: U?
LightFury01: how you guys doing
LightFury01: :)
RaunakCodes2625: :)
LightFury01: me c#
RaunakCodes2625: OO
Vaibhav0705: ik
Vaibhav0705: r
mzbear: i maintain a table of decisive board information, such as which subboards contain immediate local win for each player, and which subboard wins lead to immediate global win for each player
LightFury01: nada
mzbear: and with help of those, i can calculate if a move guarantees loss within 2 moves
mzbear: or if entering a subboard guarantees a loss
LightFury01: :)
mzbear: ohwell, since i just fixed a major bug, time to throw this into arena and see if i can get out of silver league without search algorithm ;D
Astrobytes: A heuristics-based UTTT bot eh? I don't know whether that's insanity, bravery, or a healthy dose of both! :D
mzbear: if i get into gold league, i gotta start doing better heuristics
mzbear: right now i'm only doing fairly absolute calculations. but they only take 0.024ms per turn
Astrobytes: I'm guessing you'll definitely need some kind of search if you get out of silver
mzbear: yea, this plays kinda passively really
Astrobytes: Your board representation seems nice and fast, shouldn't be any trouble to chuck a MCTS in there
Westicles: Should be easy to tune in the web editor!
Astrobytes: :D
jacek: neat, MSmits finally submitted oware bot
mzbear: the core of the bot is iterating through all valid moves, picking one with reservoir sampling, and skipping moves which are guaranteed to lose. after that it runs another two loops to try to disrupt enemy and to grab subboard wins for itself if it wants them
Astrobytes: Oh so that's what he's been quietly doing jacek
mzbear: i'm a bit upset that an earlier buggy version of my bot actually got fairly high in silver league
jacek: and without the bug its worse?
mzbear: yep. except that i seem to win against the boss 100% of the time ... if only i could get to fight it
Astrobytes: Standard CG bot writing issue right there.
Astrobytes: Boss is just the gatekeeper, it's the rabid hordes below it you gotta worry about
mzbear: oh, right, i dont set a random seed so my bot is fully deterministic ... and so is the boss
mzbear: earlier randomness was happening because i was reading uninitialized memory :D
Astrobytes: hahaha. Well that's one way to do rng I guess
niStee: :D
mzbear: oh great, here we go, now it's losing against the boss every time again ... it was just winning with that one seed ;_;
mzbear: yeah, this bot sucks. i just really REALLY dont want to write another MCTS right after Blocking
Astrobytes: Hey, you picked UTTT :D
mzbear: i didn't read the full rules of the game first, i thought it was going to be more chill ... but that thing about being sent to a specific subgrid .... argh
Westicles: I tried to warn him :)
Westicles: You could try the optims intstead, they are more fun than the bots
Astrobytes: Or make a game ;)
jacek: :notebook: :soccer:
mzbear: making a game might actually be fun, yes
Gilad-Gur-Andelman: I think there should be an extra prize for someone who wins shortest mode in JavaScript where the others write in Python
MSmits: hey jacek
MSmits: did you see this:
MSmits: http://cgstats.magusgeek.com/app/multi-oware-abapa/msmits
MSmits: making progress, but not there yet, your bot is so hard to beat :)
Astrobytes: morning MSmits :wave:
5PID3RH7CK3R: hey
5PID3RH7CK3R: Ho
5PID3RH7CK3R: Anybody?
5PID3RH7CK3R: Dead Chat
5PID3RH7CK3R: F
5PID3RH7CK3R: fef
5PID3RH7CK3R: http://chat.codingame.com/pastebin/c7d4ca4c-ba96-4432-94b0-114e645b000b
5PID3RH7CK3R: OOO noice
Astrobytes: You want kicked?
5PID3RH7CK3R: nah
5PID3RH7CK3R: im just testing
5PID3RH7CK3R: lol
Astrobytes: Well stop it then.
5PID3RH7CK3R: kk
jacek: 50%? :scream:
mzbear: oh yeah, sending enemy to play in subboards that cannot contribute to victory anymore seems quite effective in UTTT silver league, wins against the boss half of the time too. but only got rank 7, mmh mmh mmh
MSmits: morning Astrobytes
MSmits: jacek it is like 70 with you as p2, and 30 with you as p1
MSmits: your p1 game is really hard to beat
MSmits: 70 took a long time, but is more doable
MSmits: (p2 i mean)
MSmits: think there's an early mistake somewhere in your p2, but hard to be sure
MSmits: from doing a lot of meta mcts, it does seem like the bot almost does perfect play
jacek: so if i rewrite the bot to rust, you are dead
MSmits: nah, i dont need massive cg bench, it would just mean i submit twice a day to get some games
MSmits: i run cg bench to first loss, analyze the game, adjust meta mcts to search
MSmits: my bot doesnt make many mistakes either
MSmits: so i just need to find them
jacek: nice of you being nonNN at the top
Astrobytes: that was the objective iirc ;)
MSmits: without book it's a clear 5th
MSmits: big gap on both sideesw
MSmits: sides
jacek: big gap between top 5 and rest
MSmits: yes and big gap between 4 and 5 i think, well depending on submit
MSmits: sometimes i would get near 4, but i think it's lucky
MSmits: btw agade has a weird bot
jacek: and scary avatar
MSmits: one move in some cases is difference between 0% and 100% winrate
pu55ython: titties
jacek: oO
MSmits: sometimes i am adjusting my meta mcts vs you, and my agade drops hard
MSmits: because it's super deterministiv
MSmits: +c
MSmits: basically means finding a new playline vs agade
MSmits: pretty easy because it always plays the same moves
MSmits: he should do the +random thing
MSmits: so should robo, also pretty easy, but his playline does not compete with yours so it doesnt change
MSmits: btw if you were wondering:
MSmits: Moves Added: 1473
MSmits: that's all
MSmits: room for 100k more :P
MSmits: i think i run out of HD space before i run out of book space
jacek: how deep it is
VizGhar: What are you 2 talking about? You should probably open closed chat :)
Astrobytes: Nah, this is the daily sermon from MSmits
Astrobytes: (well, used to be daily ;) )
VizGhar: :D good to know
MSmits: it depends jacek. Some cases 90 plies, on your p2 line. The thing is, even though you often have 4 moves to choose from, 3 moves are so bad that no bot will pick them anyway. So branching is effectively 1
MSmits: VizGhar dont interrupt the sermon ! :P
jacek: do you have more gray hair now?
MSmits: mmh not really, less hair maybe
VizGhar: ok, ok... jeez... :D
jacek: http://chat.codingame.com/pastebin/8063c879-ea5f-4e7b-a2aa-592d5169909d
MSmits: the interesting thing is that many consistently won games go 200 plies deep, but i just need a little help preventing mistakes in the first 20 to 40 plies
Astrobytes: VizGhar: Smits is master of opening books and Destroyer of NNs
MSmits: I try :P
MSmits: yes this guy is like me jacek
Astrobytes: lol
MSmits: so my school is closed till 19 jan =/
MSmits: lockdown
MSmits: will be doing a lot of online teaching again
Astrobytes: Ah. Well, better for your health prospects I guess.
MSmits: yeah, i have some colleagues who caught it
Astrobytes: Also, more CG time :smiling_imp:
MSmits: that for sure, i just got out of a very busy work/study period
Astrobytes: Ah crap, I hope they've recovered well
MSmits: i think so, 1 still sick but i expect he will be ok
MSmits: I have to create a game and game design document for a class (as a student)
Astrobytes: Was just gonna ask about your studies
MSmits: I was thinking of recreating yavalath in unity and add some features for improving play
MSmits: like hints and such like they have for chess games
Astrobytes: Nice idea!
MSmits: its better than the alternative, they wanted me to try and create a variation of the roll a ball tutorial
MSmits: but that seems boring
Astrobytes: Why stop there though. Ultimate Yavalath. Many boards. Many dimensions :P
Astrobytes: That seems excruciatingly boring tbh
MSmits: sure i might do some other variation
jacek: roll a ball?
MSmits: its a basic unity tutorial
MSmits: move a ball, gather coins
jacek: variation would be roll a cube
MSmits: pretty good as a tutorial, but not very motivating to work on
MSmits: lol
Astrobytes: haha
Astrobytes: roll a ball, every time you grab a coin a Yavalath board pops up
MSmits: good one
MSmits: but i am allowed to just create a different game
MSmits: they wanted to make it easier for since it is only a 3 credit course
Astrobytes: Yeah, I'd go with your idea. Won't take you long, you can spend more time polishing it
MSmits: yeah, btw, i think you can even sell a yavalath game as long as you dont call it that
MSmits: the name is trademarked
Astrobytes: Lavayath. Just add a volcano somewhere.
MSmits: you cant copyright simple game mechanics apparently
MSmits: brilliant
Astrobytes: :grin:
MSmits: maybe in the middle, it's a solved move anyway
Astrobytes: Good idea!
MSmits: might have to increase the size of the board, because it might cause a lot of draws, maybe not though
MSmits: btw
MSmits: i did the playground tensorflow thing
Astrobytes: Have to test it first. Either that or you could just make the whole board lava, pieces jump around on rocks
MSmits: for my course at work
Astrobytes: ah right, how did that go
MSmits: cut it into little pieces and slowly grew the app to teach them about NN
MSmits: it worked great
MSmits: much better than i thought it would, it's really a nice thing they made
jacek: playground? you mean that graphic thing?
Astrobytes: Awesome. Your students responded well to it?
MSmits: its especially cool how you can solve some problems with more input and also with more hidden layers
MSmits: jacek yeah
Astrobytes: brb, need coffee
Uljahn: did you use specific dataset?
Uljahn: like MNIST
MSmits: https://playground.tensorflow.org/
MSmits: all 4 data classification sets
MSmits: the spiral can only be done with hidden layers
Uljahn: ah, that thing
MSmits: you dont even need to be able to code to learn about NN
MSmits: in depth even, because it really shows what hidden layers do
jacek: the pinnacle of machine learning - throw new more hidden layers and see what happends
MSmits: yeah, it also shows that if you throw in too many, your nn wont converge in any reasonable time
MSmits: kinda gives you a feel for the kind of thing researches deal with
MSmits: the trial and error
MSmits: still feel a bit lacking in knowledge because I havent made my own NN yet, but it's good enough to teach these basic things
jacek: it really kicked when i saw xor example in python, then rewritten to withtout numpy, then rewritten to c++
jacek: with 1 hidden layer and arbitrary hidden nodes
jacek: thats why im strill struggling with convolutional NNs. i could simply do complicated architecture in python using lib, but to understand it fully i'd have to write it myself from scratch
Astrobytes: well, experience is a great teacher
MSmits: yeah
ian173: miss me with that poorly written coding question
jacek: hm?
Lks10: which game you all talking about?
jacek: oware
SupSon: oware uses nn?
Astrobytes: several of the top AIs yeah
jacek: except for the 3rd
jacek: ;f
Astrobytes: indeed
SupSon: tensorflow?
darkhorse64: I am looking for N-tuple reference (either toy implementation or paper for noob). Any pointers
darkhorse64: ?
jacek: this? https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.108.5111&rep=rep1&type=pdf
jacek: SupSon there are no NN libs for python in CG.
MSmits: is it fair to say there are no NN libs in CG?
MSmits: at all?
jacek: maybe
darkhorse64: I read this one but it does not really describe what is a N-tuple or I did not understand
darkhorse64: Thanks anyway
MSmits: a ntuple is a pattern with a parameter
MSmits: when you eval you check for patterns on your board
MSmits: N stands for the number of squares
MSmits: or hexes or whatever
MSmits: in othello you can go up to 9 or so
MSmits: because there are so many possible patterns, you get a huge load of params, which you train
MSmits: using TD learning for example
jacek: well i know several papers of using N-tuples, dunno if they are more explanable
MSmits: darkhorse64, basically, you create a giant lookup array
MSmits: you pext your board to extract sets of patterns for features
MSmits: then do the lookups and calculate your eval by adding them up linearly
MSmits: correct me if i am wrong jacek
jacek: http://www.cs.put.poznan.pl/wjaskowski/pub/papers/Szubert2014_2048.pdf for 2048, https://ieeexplore.ieee.org/document/6932907 for connect4
jacek: basically its correct, MSmits
MSmits: thought so, all this seems easy to me, it's mostly the learning I am vague about
SupSon: do you use frugally deep?
MSmits: whats that?
jacek: so, no :)
jacek: i googled its some c++ nn lib
DomiKo: :joy:
SupSon: how do you encode the state of the board?
MSmits: for othello it's easy, 64 squares
darkhorse64: I have read all of that by googling this afternoon but I do not seem to be able to connect all the dots together. How to extract the pattern across the board, how to train, what are the relevant data structures ? Actually most papers require some prior knowledge that I haven't yet
MSmits: 1 bit each
MSmits: two boards
MSmits: darkhorse, use pext to extract
SupSon: no for oware
MSmits: did you use pext for yavalath darkhorse64?
MSmits: it's the same thing
jacek: i use 12 ints for houses and 2 ints for score
SupSon: oh so not one hot?
MSmits: SupSon i use a single uint64_t for the entire board
Astrobytes: ('cause he's nuts)
SupSon: for input to the nn?
jacek: SupSon oh you meant the input for the NN?
MSmits: Astrobytes yes :P
SupSon: yea input to the nn
MSmits: darkhorse64, this might help:
MSmits: https://stackoverflow.com/questions/48394866/how-to-replicate-pext-parallel-bit-extract-with-basic-bitwise-ops
jacek: then yes, i use one-hots. each house has 1 possible state of 0-31 and each score also is 1 possible state of 0-25
MSmits: or https://www.felixcloutier.com/x86/pext
darkhorse64: I know pext, the real issue is the mask (I mean which data to extract)
jacek: so i have always active 14 inputs out of possible 436 inputs at any given time
MSmits: ohh that depends on the game
MSmits: Othello uses mostly patterns connected to the corners
MSmits: like 3x2
MSmits: or 3x3
RoboStac: I just directly pass 16 floats in for my nn - 12 for the houses, 2 for the scores and 2 0's so my input layer is a multiple of 8
MSmits: so you extract those bits and look them up
MSmits: hi robo :)
RoboStac: hi
SupSon: alright thanks
MSmits: RoboStac, if you want to beat my bot again in oware, add some random to your move selection
MSmits: +0.01 or something
MSmits: might also help beat other NN's if your bot is strong enough
RoboStac: I'm not sure how much difference that'd make, it's usually very certain on the value of a move
MSmits: I'm just solving your entire playline now =/
MSmits: hmm ok
MSmits: but the game is long
MSmits: you say usually, but that's 100 choices or so
MSmits: it only takes some variation to frustrate a booker
MSmits: just an idea, make it a little harder for me
Astrobytes: wait for the chess game, you'll love that MSmits :P
MSmits: your bot is otherwise very very strong
jacek: but its 960
Astrobytes: exactly
MSmits: whats that?
Astrobytes: "love" it
MSmits: 960?
Astrobytes: Fischer Random
MSmits: ohh ok
RoboStac: a chess variation designed to beat opening books
MSmits: nice
RoboStac: 960 starting layouts
MSmits: i would not mind more games i cant book, there's too many games to book as it is. I only have so many cpus :P
Astrobytes: hahaha
MSmits: still have to do checkers and breakthrough
MSmits: havent started on those
MSmits: i kinda want to improve bot before i do
Astrobytes: too busy with raic for cg atm
MSmits: meta mcts sucks with bad bot
MSmits: ah hows that going
jacek: not if you allocated enough time for mcts...
Astrobytes: Pretty good, qualified for round 2 at least
MSmits: nice
Astrobytes: game is pretty cool
MSmits: jacek thats true, but it would go off on a bad branch for days. Better to avoid that. Also helps to have a strong solver to detect early loss/win
Westicles: there is this libtensorflowlite_c-linux64.so
Westicles: Looks like it is for dart...
jacek: oO
Westicles: and also rust
RoboStac: MSmits - looks like you were right, a little randomness does beat your opening book
jacek: :scream:
YodaMaster123: Neural network for Oware Abapa?
Astrobytes: Yep.
YodaMaster123: About that, has anyone attempted to modify aged`s neural network chatbot for another server?
YodaMaster123: *Agade.
Astrobytes: not that I'm aware of, but perhaps?
NguyenVinhHien: what is avantage of Java and Nodejs for back-end programming bros?
MadKnight: YodaMaster123 u wanna turn him into discord chatbot ?
Uljahn: shouldn't it be retrained on relevant chat logs first?
Uljahn: imagine it pinging euler on another server xD
Astrobytes: lol
LastRick: Anyone out there beat Escaping the Cat? How far off is this last test? https://www.codingame.com/replay/516134139?f=290
surperMM: Single possibility
jacek: AutomatonNN how many hidden nodes do you have
AutomatonNN: why is this contest enough to get a secret to make the game and simply have the best path to test ra
MSmits: RoboStac I knew tha :)
MSmits: by making your bots more random, I make my book more general and better
MSmits: just takes longer to "train" is all
jacek: so it was the plan all along
MSmits: yeah.. my evil plan
MSmits: :smiling_imp:
Uljahn: LastRick: as a reference, here is the last testcase with some dark trigs magic :)
https://www.codingame.com/replay/516135841
MSmits: need to share i think?
MSmits: I cant view your replay
MSmits: Uljahn
Uljahn: :confused:
MSmits: if you run a replay in the IDE it's private
MSmits: unless you share it with the share button
MSmits: thats why you're also not able to share CG bench replays
Uljahn: https://www.codingame.com/replay/516138224
Uljahn: i used the share both times :(
MSmits: :scream_cat:
MSmits: what weird stuff is that lol
jacek: the pool should be hex
MSmits: well anything is better with hex of course... but currently i am just wondering what I am looking at :P
Uljahn: that was quite overengineered solution though
MSmits: o right, i can click play
MSmits: is this a puzzle ?
MSmits: says best score up top, wondering if it's an optim, but then there's no leaderboard, so..
MSmits: :confused:
Uljahn: just a puzzle with a nice visual
MSmits: ah ok, yeah it does
LastRick: Oh wow. Zigzag? Very impressive.
LastRick: BTW I just solved it, found the problem was I was using the cat's current location to solve where on the pool edge I should go, rather than the cat's location on the NEXT step. That literally was the difference.
MSmits: ahh that's a common mistake in many problems like that
WillRises: Anyone in the chat rn that can help me with JavaScript puzzle on codinggame.com ?
I don't understand what it's asking me and feeling dumb
LastRick: Feel free anytime to ask a question about a puzzle in the chat. Just be as specific as possible.
MSmits: include the link to the puzzle
MSmits: that way people can click on it, quickly see their own solution if they solved it and maybe give you a hint
eulerscheZahl: what's codinggame.com?
eulerscheZahl: oh right, it redirects here
LastRick: Thats your one question today, Euler, youre done
eulerscheZahl: as codeingame.com does
MSmits: it's actually the real address eulerscheZahl, you just suck at spelling :P
eulerscheZahl: oh :D
eulerscheZahl: then they sent me the wrong merch
MSmits: merchandise?
eulerscheZahl: yes
MSmits: or they just know you can't spell
MSmits: is this T-shirt for Euler? Remove the G please
eulerscheZahl: https://img.printfection.com/18/21271/bT7r1IaOxZpPl8N/Next+Level+3600+-+CodinGame+Rubiks+Cube+-+Black+%28Front%29.png
eulerscheZahl: they didn't even add a separation between 2 blocks
LastRick: Wow, i didnt see that at first
MSmits: mmh ok
MSmits: why a rubiks cube anyway
MSmits: it's another nerdy hobby, but not really related
MSmits: might as well have used a chess board
eulerscheZahl: why robot arm?
WillRises: https://www.codingame.com/training/medium/shadows-of-the-knight-episode-1
Where do I start with this puzzle?
I'm new to code and looking for projects to practice on.
eulerscheZahl: because it looks cool
MSmits: I guess so :)
VizGhar: This one is fun... but it simply looks like picking buch of correct if elses hmm
https://www.codingame.com/contribute/view/5638cd2b181128c321c5e5ca1e7cf76ce9b0
struct: WillRises binary search
MSmits: WillRises you use binary search
MSmits: damnit struct
VizGhar: WillRises check forum and apply binary search :)
WillRises: Got it. thanks guys
LastRick: MSmits: you used two extra words, its your own fault
WillRises: https://www.codingame.com/learn/binary-search
MSmits: WillRises basically, you need to know the minimum and maximum coordinate of the possible locations of the bomb and jump between them
MSmits: that minimizes the number of jumps needed
MSmits: LastRick correct
WillRises: Right I got that. I just don't know how to apply that in the (terminal?).
MSmits: allright, I assume you know the current location of batman
MSmits: if the input says the bomb is to the right
Astrobytes: in the Bat Cave
MSmits: you take the currentX
MSmits: and the maximum X
Astrobytes: just off the phone with him
LastRick: Hah
MSmits: and go between them. Just average them
jacek: or maybe hes asking how does input/output work on CG
YusufEmad04: print("Hello World")
MSmits: input works as reading input for your language, depends
WillRises: Hmm I might be learning how to run a marathon here before learning to crawl
MSmits: output is basically just your languages print statement
DomiKo: hi struct
MSmits: you might be a talented baby, who knows
struct: hi DomiKo
DomiKo: You know you got bug too?
struct: WillRises you might want to start with easy puzzles
struct: DomiKo which bug?
DomiKo: in STC
eulerscheZahl: oh, you are in top20 already :o
WillRises: struct. got any suggestions on "easy puzzles"?
eulerscheZahl: i think you should stop now
struct: I saw the replay, but no bug
MSmits: some easy puzzles are harder than batman though, they're not all categorized properly
DomiKo: euler yea
DomiKo: top20 is hard no
DomiKo: the gap is so biggg
struct: do the decent, power of thor, horse racing, temperatures
MSmits: the descent
WillRises: Word. Thanks.
MSmits: those are quite good yeah
eulerscheZahl: do the descent decently
struct: DomiKo what was the bug? :|
MSmits: also dont give up when you fail the first one
jacek: :notebook: :soccer:
eulerscheZahl: :door:
MSmits: some might suit you better
WillRises: Never give up. Never surrender
DomiKo: the bug is losing :wink:
MSmits: right
struct: oh
MSmits: jacek it's still on my list. Maybe if you didnt make your bots so strong in other games, i'd already be doing the soccer game :)
struct: Maybe when I finish STC that bug is fixed
LastRick: All of the easy puzzles are definitely not created equally. "Easy", "Medium", even "Very Hard", these are subjective terms not defined by, say, a dictionary.
eulerscheZahl: is STC on your list too smits?
MSmits: smash the code?
eulerscheZahl: yes
MSmits: i did start it once, but never finished the bot
MSmits: something happened, probably a contest or a fun multi
Westicles: RIP another rockstar
MSmits: i also still need to finish my D&B
eulerscheZahl: they are dying like flies
LastRick: rockstar is (was?) a user, correct?
Westicles: ya
eulerscheZahl: i have 5 now unnamed followers in the last 3 weeks
eulerscheZahl: i think that was all rockstar
MSmits: what are you doing to them?
eulerscheZahl: ban
LastRick: what is going on?
eulerscheZahl: long story
LastRick: ahh
LastRick: how can there be so much drama on a site full of nerds coding pods
Westicles: I made a nice clash for him... oh well
MSmits: there's not that much drama i think, considering this chat is almost completely open
mzztin: poggers
MSmits: problem is mostly that teachers have their class login here and do stuff
MSmits: which is cool, but problematic sometimes
eulerscheZahl: there's a much bigger drama event in the coding world: hacktoberfest
eulerscheZahl: imagine it like a school class released on the CG chat. but instead of 1h it lasts an entire month
MSmits: fun
YusufEmad04: hi
eulerscheZahl: nonsense pull requests just so you complete 5 PRs and get a free tshirt
MSmits: maybe I am immature but sometimes their jokes make me laugh.
eulerscheZahl: :D
LastRick: had no idea about the classroom assignment stuff. But it makes sense
eulerscheZahl: some teachers park their class here. which is fine as long as they are supervised
LastRick: hahah
MSmits: I've only rarely had specific students come online here, not just whole classes
MSmits: it just doesnt work, most students cant code well enough, so they'll just mess around
eulerscheZahl: ramdeath has better students than you
LastRick: i'll say this. I enjoy the chat and site more during this time of day rather than, say, eight hours from now. Totally different.
MSmits: no doubt, different kind of class I bet
eulerscheZahl: in 8h Europe is asleep
MSmits: LastRick it's not the same every day though
MSmits: but i get your point
MSmits: eulerscheZahl is asleep
MSmits: not all of us
eulerscheZahl: :P
MSmits: well... we *should* be
Astrobytes: Smits is up all night cooing softly to his meta MCTSs
MSmits: but I doubt i'll sleep early these next few weeks
MSmits: huh
MSmits: i thought all my cameras were off
Astrobytes: :D
MSmits: did MadKnight tell you this, he's always hacking me
Astrobytes: hahaha
MSmits: ohh i got kicked off nr 1 othello
MSmits: some guy named oldjohn evaded my book
MSmits: mmhh
Westicles: easy come, easy iago
MSmits: he beats dbdr sometimes too
struct: 40 moves book lol
Astrobytes: jeez westicles that was awful
struct: https://www.codingame.com/replay/516146459
Astrobytes: :D
Westicles: heh heh
MSmits: struct that is going to make it really easy to fix for me
struct: yeah
LastRick: Westicles: :neutral_face:
MSmits: well I'll get on that then
mzbear: rewrote my heuristic engine, misplaced one ! and it took forever to find the issue, argh
MSmits: happened to all of us
MSmits: sucks
MSmits: sometimes i just place a ! because I am really excited about my code without meaning to negate it
mzbear: :D
Astrobytes: :rofl:
Maxim251: what a nice game
LastRick: :laughing::skull:
RaunakCodes2625: hello
MEtal221: hello
jacek: oh im 1st in o
jacek: and his book is really deep
OldJohn: Only 10 lines
jacek: oO
OldJohn: I took a 30 years old program
MSmits: well mine is only 1 line
MSmits: it's a very long line
jacek: logistello?
OldJohn: No but I was working with Michael
MSmits: is that one of the designers ?
jacek: :tada:
jacek: michael buro?
Astrobytes: Obviously Michael Buro
OldJohn: Yes This was my program I took
jacek: nice place, considering ai advecements in the last 30 years
MSmits: nice to meet you OldJohn
MSmits: Maybe you know Renard then, he also used a 30 yr old othello program afaik
OldJohn: Yes here is a human evaluation no bitboard hastable and endgame
OldJohn: I know Sylvain yes
MSmits: oh thats his real name then?
OldJohn: In fact we did not see each other for a long time
MSmits: it's cool to meet up here then
jacek: not even n-tuples?
Astrobytes: It's like CG has an underground computer board game expert network
Astrobytes: :)
mzbear: grr. i cant make a heuristic that beats the boss AND all others to reach the boss. i give up, i'll start writing a search algorithm after all, i can't do this with pure heuristics :(
MSmits: seems like that yeah
MSmits: mzbear you dont have to
Astrobytes: tric too, checkers was his thing iirc
MSmits: if you have a heuristic that beats boss
MSmits: you will keep submitting
MSmits: and push the other guys above the boss and promote them
MSmits: when they're all gone, you got it
OldJohn: I think dbdr program is probably the best at Othello from a Othello player point of view
mzbear: MSmits :DDD
mzbear: that's certainly ... an interesting strategy ...
MSmits: OldJohn possibly, but dbdr did not add random
MSmits: it's quite easy to counter
MSmits: jacek much less so
Astrobytes: He used multi probcut if I remember rightly
MSmits: yeah he went all out on this one
jacek: but mine loses to dbdr considerably
jacek: im 1st due to submit luck
MSmits: right and also because my bot wins 96% vs dbdr and maybe 80% vs you
MSmits: because of the damn random
MSmits: i did a lot of submitting after which the game went still, so thats also why you ranked above dbdr
MSmits: not much luck to it
OldJohn: The long opening line versus your program is only because my program did not find the winning move but I saw it and check with more thinking time with my computer
jacek: ohh so personally you are good othello player?
MSmits: that's cool OldJohn, I sense a kindred soul
OldJohn: Are you all using bitboard ?
Astrobytes: Great to have guys like you playing on here, really
Astrobytes: I believe mostly everyone is using bitboards yes
OldJohn: In fact I have a bad program in '89 that was using bitboards that competed in Computer Olympiads
jacek: most here use bitboards
MSmits: yeah I use a 2x 64 bit bitboard
darkhorse64: OK, I think I understand how n-tuples works (at least how to implement it, why it works is another story). Still have to work on TDL namely how to update the weights
MSmits: thats where i am also
MSmits: the patterns work because in othello (if this is the game where you do it) local patterns can tell you the value of the board.
MSmits: you wouldnt be able to do this in uttt, because moves will make you jump all over the board
MSmits: in othello you will have some isolated fronts
MSmits: not entirely.... but close enough
darkhorse64: It reminds me of UTTT tritboards for the encoding. It might also be applicable on Y with 4-tuples
MSmits: thats exactly what i do in uttt, i use tritboards for lookup. I guess it's kind of like using miniboards as ntuples
MSmits: but as eval its not that great
MSmits: because of the interdependence of the miniboards
MSmits: it's really a domain specific thing
darkhorse64: I have not touched that thing for months. Even if I know implementing bonus for teccles, mini board wins or free moves would help
struct: any other game anyone would like to see?
Astrobytes: Don't forget Amazons
MSmits: is gomoku worth doing?
MSmits: I'm going to regret suggesting this :P
MSmits: not sure if maybe it's solved or unbalanced or anything, i came across it reading papers
MSmits: apparently go players play it to relax after a go game
MSmits: I dont like go, but i do like to relax
Astrobytes: stops them go-ing moku
MSmits: lol
MSmits: isnt amazons solved?
MSmits: probably you can grow it to a larger size
MSmits: or something
eulerscheZahl: i'm pretty sure Nine men's morris would be a boring game here on CG
Astrobytes: need 12+
struct: I dont think Amazons is solved
MSmits: nine mens morris is easily solved
MSmits: could also be grown
Astrobytes: that's why I said 12
Astrobytes: can't remember much about it though, I read some papers
Astrobytes: a while ago
Astrobytes: afk for a bit
struct: seems like gomoku is forced win/draw for black
struct: They use swap rule
MSmits: is the game interesting when swapping?
MSmits: because the same is true for yavalath, its a forced win for p1 on 7 starting hexes
MSmits: but with swap its a very interesting and balanced game
struct: Its a different kind of swap
MSmits: oh?
struct: "In Swap2 rule, the first player starts by placing three stones (2 black and 1 white, if black goes first) on the board. The second player then selects one of three options: play black, play white and place one more stone, or place two more stones and let the first player choose the color.[5][6] This is essentially a slightly more elaborate pie rule."
MSmits: ahh ok
reCurse: Ohh I remember nine men's morris. It was to study minimax in some AI course I had at university decades ago.
reCurse: They organized an AI tournament for it, little did I know at the time it would be useful so long after ;)
eulerscheZahl: sounds like you are older than i always thought
reCurse: I know -_-
reCurse: 18 years ago so maybe decade was not quite plural yet
struct: for some reason I Thought he was in 20-30 range
struct: 25-30*
reCurse: Need to bump that up a little
MSmits: well struct you were right he was
MSmits: and then he wasnt anymore
MSmits: I always thought he was my age
MSmits: almost 40
eulerscheZahl: that matches my imaginations
reCurse: Yeah getting closer
reCurse: sigh
LightFury01: hello guys
LightFury01: how you doin
LightFury01: :)
monami7000: great
monami7000: am 56 years old, learning to code
MSmits: you're never too old to code
MSmits: I mean look at reCurse, he still does it
MSmits: :P
reCurse: Hey you're older than me though not by much
reCurse: shush
MSmits: :)
Astrobytes: getting older ain't that bad tbh
Astrobytes: For now
struct: MSmits I think i asked you this before, do you think bitboards are worth on STC?
LastRick: Even those of us that are *cough* even older can learn a lot here. This site has been a pleasant surprise
Astrobytes: Nice to hear. Good to have another convert on here. And there're plenty 'mature' people on here ;)
mzbear: ffs my bot got high enough to challenge boss ... 12 losses against the boss, no wins ... outside the arena i keep winning half of the time
LastRick: Boss: Know your place, son.
wisp1st: Know your place Boss
kovi: oh, recurse is older than me :o
eulerscheZahl: Artrosebytes
Astrobytes: lol wot
Astrobytes: how old are you kovi?
kovi: 50
mzbear: you know what, i'm gonna cheat ... i'll just predetermine some random seeds that win against the bot every time, since it's deterministic
Astrobytes: I'm 38, I don't think recurse is 50+ btw ;)
eulerscheZahl: that's now what I understood either
but a good way to trigger him
mzbear: actually, i could test out every top bot in silver league to see if they are deterministic, and figure out some seeds that win against most of them ;D
Astrobytes: lol euler
Astrobytes: write a search already mzbear, your masochistic approach is making my teeth itch :D
mzbear: there we go, 3 hardcoded seeds that each always win against the boss, chosen at random (so i wont get stuck against other deterministic ones). i feel so bad doing this
mzbear: but i want to get to gold league without search
LastRick: Good gracious, more contributions? I think we're over 110 now.
LastRick: I was going to make a joke earlier this morning about breaking through 100. Hah, we've blown by now.
Astrobytes: Yes, it's getting a little crowded in there. Mostly clashes too I think.
jacek: clashes :f
LastRick: Over 80 of the 110
reCurse: Is that the game now, triggering me? :P
Astrobytes: :/
Astrobytes: lol
jacek: nn uttt clashes?
reCurse: wtf people think I'm 50 now
Astrobytes: Yeah, GRAMPS :P
reCurse: I'm 36 geez
jacek: ok boomer
LastRick: Hey, thats Euler's nickname
LastRick: gramps
Astrobytes: I was pointing it out to kovi that you were not in fact, older than him
reCurse: Game is over, I'm triggered :p
Astrobytes: hahaha
kovi: im reliefed then (loosing vs. younger is acceptable) sorry i misunderstood the relations in chat, you refered to msmits probably not manami
reCurse: Haha yeah
reCurse: Strange curve, usually it's acceptable to lose vs older
reCurse: When does it shift?
AI_says_hi: Hey folks I have a general question: Som eof the bot programming challenges recommend using machine learning, e.g. genetic algorithm. Is the ML-mechanism supposed to learn within a single game? Or is there a way to carry over parameters? Thanks :)
elcolumbio: at 45
elcolumbio: 42
Astrobytes: 37 reCurse :P
jacek: you cant 'save' anything from the game
reCurse: F
Astrobytes: you can run an online search, or you can download replay data and train offline
LastRick: I've had a related GA question sitting on the forum for a few days with no answer, I'll throw it out here. I've been reading a lot recently about GA, mostly word guess examples and I would love to code Search Race. I know ppl aren't going to share code. But any suggestions about the choice of gene/crossover/mutations would be appreciated.
reCurse: Whatever gets you out of it :P
Astrobytes: thrust + angle?
RoboStac: LastRick - http://files.magusgeek.com/csb/csb_en.html and just ignore anything about blockers
kovi: loose vs. younger:
http://www.rhsmpsychology.com/Handouts/verbal_and_nonverbal_iq__age.htm
eulerscheZahl: https://en.wikipedia.org/wiki/Flynn_effect
eulerscheZahl: IQ is defined via average intelligence
if your intelligence remains constant but newer generations get smarter, your IQ value drops
struct: so the older people IQ might increase
reCurse: Hmm...
MSmits: struct, my bitboard floodfill was not very fast on STC, but i never finished the bot
reCurse: Not sure what to make of both :)
LastRick: Robo: I've read that article and one point I'm confused about is he's evaluating chromosomes with pod1 and pod2 directions and he picks the "best" one to output. But he also says he looks six steps into the future. Is the "best" option the one that puts in the best position for the next turn? Or the one that puts in the best position six turns in the future?
struct: I guess I won't try it then
struct: Also I'm not familiar with bitboards so I guess it would take me a while
MSmits: dont try it on STC, but do try it on any boardgame
MSmits: its really not that hard, only a few things to learn
MSmits: i should say, its not hard to get the bitboard state working. But to make a quick sim is sometimes a challenge
MSmits: for example bandas pushing and oware seeding
MSmits: othello has free code you can google
MSmits: so the sim problem will be solved there
RoboStac: LastRick - play out the 6 turns and then calculate a score (if you look at the score function in 5.2.1 you can see the loop to play turns before evaluating)
jacek: yeah, but what about boards over 8x8, or over 11x11 (so you cant cheat with int128_t)
MSmits: i did bitboards on ooc
MSmits: just make an array of ints
jacek: oO
MSmits: recurse won with bitboards on ice and fire
jacek: and i see youre resubmitting O bot now :o
MSmits: yeah, i fixed it
jacek: fixed?
eulerscheZahl: added entry to opening book
MSmits: yeah so it doesnt lose vs our new friend :)
struct: I saw puyo puyo code of a person doing it with 2 int64
eulerscheZahl: so last to edit the book wins?
MSmits: in theory yeah
MSmits: but so far that always seems to be me :P
reCurse: Saying you win with bitboards is a bit like saying you win with C++...
MSmits: sure, but jacek was suggesting bitboards are hard to use in some games
jacek: HTML is programming language
eulerscheZahl: temporarily i got first with code
reCurse: So a bit boring
jacek: pun intended?
reCurse: Maybe
jacek: MSmits you could print the book moves like him
MSmits: or i could do it like you and always print book
MSmits: didnt you do that in one game?
jacek: the other O game
MSmits: ahh ok
MSmits: I wouldnt print my book moves, because that would make it really easy to counter
jacek: i used to in bt to print book randomly up to 38..46 plies
MSmits: thats funny
jacek: to troll the boomers
jacek: bookers*
MSmits: nice mistype
Astrobytes: excellent
MSmits: typo
jacek: or was it?
MSmits: no idea
MSmits: you are a mystery
MSmits: aww damnit, i lost my 46th othello game
MSmits: damn your random
MSmits: :rage:
eulerscheZahl: :D
Skril: thanks eulerscheZahl for Space Maze ; it was a fun and interesting puzzle :-)
eulerscheZahl: wouldn't it be boring with everyone deterministic?
MSmits: yeah i am just kidding. It would be very boring
eulerscheZahl: i saw you solved it, congrats Skril
eulerscheZahl: you are the 7th to solve it without me
MSmits: good job skril, apparently that one is very hard
Astrobytes: 7 now, nice
Skril: thanks :-)
jacek: oh you lost again
MSmits: yeah, i never completely finished royale and your's playlines, too many options.
MSmits: also royale kept resubmitting different versions, i got a bit fed up :P
MSmits: doing this makes me think about what othello AI tournaments must have been like
MSmits: opening books must have been huge. Checking official games, trying to counter those moves etc.
MSmits: either that or they found some way to prevent it... but i doubt it, the game is very simple
MSmits: as in few states
jacek: thats why at some point they banned opening books for chess or forced some balanced openings
MSmits: they would have to look at the source code to know someone is using one
MSmits: a book that counters a specific bot is very small, so it's not even going to show in the size of the executable
MSmits: i guess they can *suspect* and players can be sporting enough to try not to cheat
reCurse: It's solved by forcing random openings
reCurse: Or in chess, use chess960
MSmits: yeah I guess so
MSmits: Onitama does a good job of this
mzbear: WTF .... i used hardcoded seeds that guarantee 100% winrate outside of arena ... and still lost against the bot twice in the arena
MSmits: calculation time might be somewhat different in the arena for some languages
eulerscheZahl: which game?
mzbear: UTTT
MSmits: i should say performance
eulerscheZahl: gold boss? is not deterministic
mzbear: silver boss
mzbear: i'm trying to beat silver without a search algorithm :D
eulerscheZahl: hm, i don't know that boss
MSmits: what algorithm do you use?
eulerscheZahl: but your plan doesn't sound fun to me
MSmits: if else?
MSmits: or eval?
darkhorse64: trueAI
eulerscheZahl: GAimax
Astrobytes: You need a Prune Array
eulerscheZahl: was waiting for that
mzbear: a lot of if else, practically
MSmits: well you can solve regular TTT with a reasonable amount of if else
mzbear: the only "search" thing is two step lookup towards certain loss or certain win
MSmits: ah ok
mzbear: other than that, it generates moves and does rules like always taking a local win if it blocks enemy global win, etc
mzbear: and once no rules match, it goes through the possible moves and picks disruptive moves first, i.e. ones that sends enemy to subboard that can no longer contribute to global win
mzbear: etc
MSmits: ah I see
MSmits: I can totally see getting into gold with a good heuristic bot like that
MSmits: did you add teccles?
jacek: and tentacles
Astrobytes: that might be the missing part
Astrobytes: (teccles)
tarikyildiz: hello here the puzzles are very confusing to me, what should I do?
MSmits: in case you dont know teccles: It means, when you come to an empty board, you pick the move that has the same position on the miniboard as the miniboard has on the big board
mzbear: people keep mentioning the teccles but i didn't find what it is. someone described it briefly and i got the idea of the disruptive moves from that, but my disruptive move play only triggers late game
MSmits: or in other words, that sends the opponent to the same board
MSmits: this is a very strong heuristic when employed by search bots. I see no reason why it wouldnt help for a heuristic bot
mzbear: so i understood it correctly. i didn't try it, though, because it feels a bit odd, but maybe i'll try it
mzbear: my current bot got rank 3 in silver league ... so close
Astrobytes: and teccles works better early-mid game too right? So the late-game disruptive move should kick in at the right moment
MSmits: I played millions of games locally, it is almost always the best move
mzbear: huh
MSmits: exceptions maybe in the late game (20 ply + ) but even then its almost always the best move
Astrobytes: If you add that and get legend I'll give you 0.001€
Astrobytes: Next year.
jacek: silly, there wont be next year
MSmits: err, you mean after the UK is out of the EU?
Astrobytes: Yes :(
eulerscheZahl: that doesn't stop him from the ability to own Euros
MSmits: true
eulerscheZahl: i still have some Kunas
Astrobytes: This is true.
MSmits: i dont know what kunas are
eulerscheZahl: the currency in Croatia
MSmits: is this some fantasy game currency
MSmits: oh ok
Astrobytes: I've got some drachmas hanging around somewhere
eulerscheZahl: fantasy Croatia :D
MSmits: didn't you guys have marks?
eulerscheZahl: yes
eulerscheZahl: Deutsche Mark or DMark for short
Astrobytes: now they have Merks
MSmits: right
MSmits: is a Merk a follower of Merkel?
Astrobytes: Merkel Mark = euro
MSmits: owww
MSmits: you're operating on another level today astro
Astrobytes: well, I mean I just made it up but, it works
MSmits: cant reach it
Astrobytes: you've only just come back, give it a few days
MSmits: hey i was never really gone. I just didnt say anything :P
tarikyildiz: Hello. Do you have any Turks on this site?
Astrobytes: yeah we noticed :P
MSmits: let me check tarikyildiz
Astrobytes: tarikyildiz: quite a few I believe yes
MSmits: well not on othello anyway
Astrobytes: Isn't the guy currently streaming Turkish?
**MSmits should have gone to the full leaderboard
Astrobytes: why tarikyildiz? Can we help you with something?
MSmits: There are 23171 accounts from Turkey
struct: Not many
MSmits: Highest ranked is Jazzcript at 1447
struct: only 20 on fall challenge 2020
MSmits: about the same number as Dutch players
MSmits: (23688)
eulerscheZahl: but i can tell you 3 dutch players right away
struct: but 126 dutch on FC2020
tarikyildiz: thank you astrobytes i write when i need help :)
Astrobytes: OK cool tarikyildiz :)
MSmits: which 3?
struct: you, counter and someone else
eulerscheZahl: you, jh bonarius, counter balance
MSmits: counterbalance me and ?
MSmits: oh bonarius sure
Astrobytes: what happened to Mr Bonarius
Astrobytes: ?
SBtree-bit: I can't wait!!!
eulerscheZahl: he became the wood clown
MSmits: great avatar
Astrobytes: He was kinda down on his luck I remember
Maxim251: How pasible is to make code 29 letters size? In perl and Bash on Temperatures
MSmits: I think he had problems with the short timeframe of contests
MSmits: as in 10 days is too short for him
MSmits: he liked locam
Astrobytes: Yeah that too, but like he got divorced and stuff
mzbear: hmm. teccles made the early game of my bot quite a bit more aggressive, but late game weaker against the boss :O
MSmits: Maxim251 people do some extreme stuff to code golf
eulerscheZahl: oh, i totally missed that personal drama
MSmits: some languages are more suited to it
Astrobytes: wasn't a drama as such, just remember him mentioning it when he reappeared a while back
MSmits: mzbear i would still keep it, it lowerst the branching a lot
MSmits: you can fix it with different if elses
MSmits: it's objectively an improvement, but it might not mix well with the rest of your bot as it currently is
Astrobytes: try just your disruptive move heuristic in the endgame and use teccles earlier
Astrobytes: (and variations on that)
MSmits: if the boss is completely deterministic, you can also just make a little opening book
mzbear: well, my bot kinda relies on luck :D
MSmits: just put in some states your bot encounters every game and force it to do some move you know is good
Maxim251: my code have 222 letters long, but if I replace if statment with ? : I could have few extra. All variables are 1 letter long, and I cant do nothing more, because i need keep {} () ; what are extra letters
MSmits: i dont know Maxim251, but you can google code golf tips or do a lot of clashes in shortest mode
MSmits: i picked up some tips there... though i am not that great at it
mzbear: i'll try cheesy approach first ... 3 hardcoded seeds that always win against boss outside arena, let's see how well teccles works against other players
MSmits: well you're not searching so cheesy is a must
MSmits: this is game is about as search friendly as you will find
MSmits: and heuristic unfriendly
Maxim251: This is the code
MSmits: oh i have another heuristic
Maxim251: http://chat.codingame.com/pastebin/061ae823-3895-42dd-95f5-567b04f5aa6a
MSmits: mzbear
Astrobytes: I told him, but he took the masochistic route :D
MSmits: never play middle square
MSmits: if there are less than 2 squares filled
mzbear: oh? that's interesting
MSmits: unless it's the starting move
Lks10: you can remove the indenting to save characters I think
MSmits: middle square is bad
MSmits: because you send the opponent to the middle board
MSmits: sometimes its still good, but not when you only have 0 or 1 squares filled on a board. There's better ways to get 3 in a row then
Maxim251: This one is hardcore... and have 212 letters long
Maxim251: #include <iostream>
using namespace std;int t;int a=5527;int main(){int n;cin>>n;cin.ignore();for(int i=0;i<n;i++){cin>>t;cin.ignore();if(abs(t)<abs(a)||(abs(t)==abs(a))&(t>a)){a=t;}}if(t==0){a=0;}cout<<a<<endl;}
MSmits: mzbear you can also watch some top 10 matches and mimic the playstyle
MSmits: you will see a lot of teccles then too
struct: for(int i = 0;i++<n;)
struct: is shorter
Astrobytes: Maxim251:
https://www.codingame.com/forum/t/tips-and-tricks-for-code-golfing-in-c/694
Maxim251: Oh thanks :)
mzbear: huh wtf
mzbear: so, teccles worked REALLY well against players in silver
MSmits: good :)
struct: I told you mzbear...
mzbear: but it seems so unintuitive!
MSmits: it's a pretty deep tactic
MSmits: not obvious
MSmits: the main thing is (I think)
Maxim251: #Astrobytes, thanks a lot :)
MSmits: when you win a board
reCurse: Complaining about unintuitiveness in uttt
mzbear: 50% of battles done and i'm already at first place
reCurse: Sweet summer child
Astrobytes: you're welcome Maxim251
MSmits: you want there to be as many of your opponent marks in there
MSmits: because those are wasted
mzbear: oh lol, i pulled some other guy above boss with me
MSmits: so when you send your opponent to the same board, you guarantee that he has marks on there as well
MSmits: meaning that if you win it, he wastes them
Maxim251: for(int i = 0;i++<n;) I didn't know that this is posible :)
MSmits: I think that's part of it anyway
eulerscheZahl: for(;;)
is the same as while(true)
MSmits: its hard to see. But meta mcts doesnt lie
Westicles: Maxim251, you don't need an endl
Maxim251: Yes, I already removed it when I realised that
Maxim251: Thanks
Maxim251: Code size is now 202
Astrobytes: finally getting an endl on things
Astrobytes: (I'll just get my coat)
mzbear: promotion to gold league in 25 minutes
Astrobytes: congrats mzbear!
Maxim251: Starting at using namespace, everything is on that line. Only IOstream is on seperate line
mzbear: i should've just tried teccles earlier ... i didn't do it because it didn't make sense to me. lol
eulerscheZahl: got me from bottom legend to mid legend
Astrobytes: There's usually good reason when people give you tips
Maxim251: Oh, Yes, I can remove ints, and all variables assighned to one int
mzbear: even if there's a good reason, it doesn't feel good if i dont understand it :P
struct: cout << t?a:0 << endl;
struct: and you can remove the if(t==0){a=0}
Astrobytes: as reCurse will tell you, good luck understanding UTTT
MSmits: the main problem with uttt is that in the beginning, many moves are about equally good
Maxim251: #include <iostream>
using namespace std;int t,a=552,n,i;int main(){n;cin>>n;cin.ignore();for(i=0;i++<n;){cin>>t;cin.ignore();if(abs(t)<abs(a)||(abs(t)==abs(a))&(t>a)){a=t;}}if(!t){a=0;}cout<<a;}
MSmits: there is no clear difference
MSmits: aside from trying to connect two x's or O's
Maxim251: #include <iostream>
using namespace std;
int t,a=552,n,i;int main(){n;cin>>n;cin.ignore();for(i=0;i++<n;){cin>>t;cin.ignore();if(abs(t)<abs(a)||(abs(t)==abs(a))&(t>a)){a=t;}}if(!t){a=0;}cout<<a;}
eulerscheZahl: do you need that cin.ignore()
eulerscheZahl: ?
struct: I dont think he needs it
Astrobytes: chat golf eh
Maxim251: Hm, what I can do with that?
eulerscheZahl: abs(t) == abs(a)
t*t == a*a
Astrobytes: remove it
mzbear: so, now i gotta wait 20 minutes before i get to see how badly this heuristics bot performs in gold league ... i suppose i'll take a look at that space maze that was mentioned earlier
Maxim251: Yes, It works.
Astrobytes: mzbear, you'll need a raycasting NN and spend a week writing down all the testcases
MSmits: or just learn a search, it's really not that hard for a simple game like uttt. Unless you already know search and just dislike it
Maxim251: #eulerscheZahl Thank You. I was thinking that I need it
MSmits: huh what Astrobytes?
eulerscheZahl: then we are getting in a range where std::cin might be shorter than
using namespace std
MSmits: he thanks you by giving you your own channel eulerscheZahl
Astrobytes: It was some guy we had on here for like 2 weeks going on about space maze, euler remembers
mzbear: i did fairly fine with MCTS in Blocking, i just didn't want to do another MCTS right away
eulerscheZahl: his ray casting didn't go anywhere
Astrobytes: his rays were blocked
Maxim251: #eulerscheZahlabs(t) == abs(a)
t*t == a*a
MSmits: ahh ok mzbear, makes sense
Maxim251: I dont know how aplay that to my code
Maxim251: My code now have 166. WOW
eulerscheZahl: "abs(t)==abs(a)"
you have that verbatim in your code
MSmits: nice
mzbear: i figured it would be more fun to get into gold league without search, and it was indeed fun... kinda :D
Lks10: Go for minimax on UTTT for a change
MSmits: my theory is that if you go minimax and work out the best heuristics for pruning, you can later use this in a mcts
Maxim251: yes
Astrobytes: Surprised you haven't attempted that yet MSmits
eulerscheZahl: replace it by a*a==t*t
MSmits: i did attempty a minimax
MSmits: my problem with minimax is that i always do it after i heavily optimize a mcts
MSmits: so its always worse and that demotivates me
Maxim251: Wow, It works...
Astrobytes: heh heh, not like you at all
Maxim251: !! Yes, now I know why I use abs...
MSmits: Maxim251 -3*-3 = 9
Astrobytes: you know why Maxim251?
MSmits: self multiplication removes minus signs
Lks10: I have the reverse problem where I have an optimized minimax, but dont want to work on mcts lol
MSmits: like abs does
Maxim251: This mathematics removes -.. Ohhh
MSmits: unless you use i
eulerscheZahl: and you have more abs in your code
MSmits: be careful of i
eulerscheZahl: stop flexing with your muscular belly
MSmits: :P
Maxim251: 154 now,
Astrobytes: :rofl:
Astrobytes: nice Maxim251
Maxim251: #eulerscheZahl Thank You so much, You show me greate stuff for lenght of code optymisation :)
Astrobytes: also you can use import rather than include
MSmits: it's not even his specialty
eulerscheZahl: that's creating weird puzzles
MSmits: you're not really into code golf are you euler?
Maxim251: #include <iostream>
using namespace std;int t,a=552,n,i;int main(){cin>>n;for(i=0;i++<n;){cin>>t;if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}cout<<a;}
eulerscheZahl: nope smits
MSmits: yeah you make good stuff
eulerscheZahl: using namespace std;
Maxim251: Yes?
eulerscheZahl: remove and use std::cin
MSmits: and std:cout?
eulerscheZahl: of course
MSmits: can you do some alias thingy for std?
Westicles: don't need int with main
MSmits: or is it not worth it?
Astrobytes: #import<iostream>
Maxim251: 144 :D
Maxim251: Is blowing my head now :D
Maxim251: :alien:
eulerscheZahl: if(!t){a=0;}cout<<a;
eulerscheZahl: struct shortened that already
Astrobytes: depends how many times it's used MSmits
struct: or you can even do
eulerscheZahl: for(i=0
move the initialization to declaration
Maxim251: But cout now need std::cout
MSmits: right
struct: cout<<t||a;
struct: if im not mistaken
Maxim251: #include <iostream>
int t,a=552,n,i;int main(){std::cin>>n;for(i=0;i++<n;){cin>>t;if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}std::cout<<a;}
eulerscheZahl: or make it a global var, that's preinitialized
Astrobytes: replace #include <iostream> with #import<iostream>
Maxim251: I made already global,
eulerscheZahl: for(;i++<n;)
Maxim251: Also for "for" statment is using global i
Maxim251: But i - I should have at 0 ?
struct: for(;--n;)
Maxim251: YES WORKS.
eulerscheZahl: while(--n)
for readability
eulerscheZahl: which is a killer argument when golfing
Maxim251: :thermometer_face:
Maxim251: Thats insane
Maxim251: 144
eulerscheZahl: you were at 144 before
Maxim251: Maybe because of std::cin?
Maxim251: #include <iostream>
int t,a=552,n,i;int main(){std::cin>>n;for(;--n;){std::cin>>t;if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}std::cout<<a;}
struct: if(!t){a=0;}std::cout<<a;}
Westicles: you still have the int on main
Astrobytes: ah yes, you missed it from previous version
Astrobytes: 1. Remove int from main.
Astrobytes: 2. replace #include <iostream> with #import<iostream>
eulerscheZahl: remove i
struct: std::cout<<t||a;
Maxim251: No, after removing int and adding main to ... Error
Maxim251: #include <iostream>
int t,a=552,n,i, main(){std::cin>>n;for(;--n;){std::cin>>t;if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}std::cout<<a;}
Maxim251: I must have int for main
struct: wait my thing might not work
Astrobytes: no, you missed a semicolon
Astrobytes: @Maxim251
Maxim251: Hmmm still error.
Astrobytes: works for me
Astrobytes: i, main() -> i;main()
Maxim251: #include <iostream>
int t,a=552,n,i, main(){std::cin>>n;for(;--n;){std::cin>>t;if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}std::cout<<a;}
Maxim251: :fearful::scream: Yes, works.. How that posble
Astrobytes: because you don't need to specify the return type of main
Astrobytes: now change your #include
Maxim251: Shows me requires a type specyfier, but without int works
struct: std::cout<<(t?a:0);
Maxim251: then it will behave like void
struct: if(!t){a=0;}std::cout<<a;}
Astrobytes: #import<iostream>
Westicles: Let's do space maze together next
Astrobytes: lel
struct: for(std::cin>>n;--n;)
struct: for(;std::cin>>t;)
struct: and you dont even need to read the number of temperatures
struct: you can remove n
struct: oh you need to read n
struct: im sorry
struct: you just dont need to use it
struct: for(std::cin>>t;std::cin>>t;)
Maxim251: How to write? I am lost
Maxim251: #include <iostream>
int t,a=552,n,i;main(){std::cin>>n;for(;std::cin>>t;){a=t;}if(!t){a=0;}std::cout<<a;}
Maxim251: sorry
struct: int t,a=552;int main(){for(std::cin>>t;std::cin>>t;)
Maxim251: #include <iostream>
int t,a=552,n,i;main(){for(;std::cin>>t;){a=t;}if(!t){a=0;}std::cout<<a;}
Astrobytes: And once more (last time):
replace #include <iostream> with #import<iostream>
eulerscheZahl: you need a macro for that astro
Maxim251: int t,a=552;int main(){for(std::cin>>t;std::cin>>t;)
Maxim251: not working
struct: that is not the entire code obviously
struct: Its just the start before the { of the for loop
MSmits: you guys are really on a mission to help this guy :)
MSmits: it's nice
mzbear: but if he's just copypasting everything blindly, you're not helping him :D
Astrobytes: Indeed eulerscheZahl :D
eulerscheZahl: it's not about helping anymore
eulerscheZahl: we are golfing
Astrobytes: it's about golfing
Astrobytes: lol
MSmits: aha
mzbear: my c++ solution to that is 106 chars, and there might still be room to trim a few easily
eulerscheZahl: for(...){instr1;instr2}
for(...)instr1,instr2
eulerscheZahl: removes the {}
mzbear: for example, that 552 ... i thought the specification said 5526 so i initialize my variable with 9999
eulerscheZahl: int n,r,t;main(){std::cin>>n;r=n*1e3
eulerscheZahl: my code
eulerscheZahl: and print r at the end
Maxim251: #import<iostream>
int t,a=552,n,i;main(){std::cin>>n;for(std::cin>>t;std::cin>>t;){if(t*t<a*a||(a*a==t*t)&(t>a)){a=t;}}if(!t){a=0;}std::cout<<a;}
//for(std::cin>>t;std::cin>>t;)
Maxim251: found -2 expected 1
eulerscheZahl: now you added the variable i again
jacek: oh my
struct: (╯°□°）╯︵ ┻━┻
Astrobytes: at least I don't need my macro now
struct: #import<iostream>
int t,a=552,n,i;main(){std::cin>>n;for(std::cin>>t;std::cin>>t;)
struct: #import<iostream>
int t=0,a=552;int main(){for(std::cin>>t;std::cin>>t;)
eulerscheZahl: why did you add the *int* main again?
struct: I initialized t to 0 by mistake
**eulerscheZahl slaps struct around a bit with a large fishbot
struct: oh lol
Astrobytes: :D
Maxim251: sorry, somebody else paste it previously, then I update code, but I make mistake and I go back to working state..
mzbear: hmm, i'm actually not losing every battle in gold UTTTT ... rank 423 out of 631
jacek: gold uttt? nice
Astrobytes: good stuff
mzbear: gold uttt ... without a search algorithm
Maxim251: Yes I see it t=0;
struct: you can to t;
struct: or t,
MSmits: gj mzbear
MSmits: btw these heuristics you used, actually went unnoticed on uttt for a year or more after the multi went live
MSmits: i mean teccles and such
MSmits: so you're pretty advanced already :P
Maxim251: Yes, works, is 5 letters shorter with
Maxim251: for(std::cin>>t;std::cin>>t;)
Astrobytes: yeah, when you do a search you should be destined for greatness mzbear
MSmits: mz:bear:
Maxim251: 133 now
struct: exec("...");
MSmits: gonna take out the trash, i expect you to be below 130 when i get back
Maxim251: My head is now hurting me.. Thats insane how I could reduce code...
eulerscheZahl: you alone :)
mzbear: well yea, the entire AI of mine takes 0.01ms - 0.05ms per turn to do its job
Astrobytes: :smirk:
Maxim251: No, never, Thanks To You ALL
Astrobytes: hey, we had fun
Westicles: please add a comment attributing us all
Astrobytes: Just don't ask us to do that again :P
Maxim251: My self I will take 2 years after I will find that insane tricks
eulerscheZahl: we didn't even start yet
Maxim251: I will try, But thats ... Oh I need print that on paper an memorise it :f
Maxim251: :D
Maxim251: :nerd::hugging::ghost:
eulerscheZahl: the whole check if you should update the temperature can be done with a polynomial instead of if(t*t<a*a||(a*a==t*t)&(t>a))
eulerscheZahl: btw you can remove some () here
Astrobytes: polynomial: when you don't feed your parrot.
Maxim251: Ill check that
Astrobytes: (sorry, my maths teacher used that in school)
eulerscheZahl: i don't get it
eulerscheZahl: not sure if a language barrier or just stupid
Astrobytes: Stereotypical name for a parrot = Polly. Don't feed it? Polly no meal.
Maxim251: I think I can remove {} from if statment
eulerscheZahl: ah :D
jacek: if you teach parrot to say "parrot, parrot", youll have a pokemon
Maxim251: 129
eulerscheZahl: https://en.wikipedia.org/wiki/The_Adventure_Series
but the parrot is Kiki and Polly is the aunt
Astrobytes: Oh jeez, Enid Blyton
Maxim251: :thermometer_face:
LastRick: Jokes and idioms are always the toughest to translate
eulerscheZahl: read 10 books and you pretty much know them all
Astrobytes: read 1 Enid Blyton book and you know them all
eulerscheZahl: but child me liked them
Maxim251: 125
Astrobytes: I only read a few since my grandmother had a few of them
Astrobytes: *only read them
Maxim251: #eulerscheZahl I fount that, I know what you mean to remove () from (t*t<a*a||a*a==t*t&t>a)a=t;
Maxim251: 123
Maxim251: 119
eulerscheZahl: if(condition)doStuff
!condition?:doStuff
eulerscheZahl: #fr just pinged me
eulerscheZahl: sokoban :imp:
Astrobytes: I forgot you made that
DomiKo: omg
DomiKo: I didn't know
DomiKo: it is finished
DomiKo: I have to do it!
Astrobytes: lol
Westicles: hah, they should kick him for english
Maxim251: !t?:a=0; hm I think I did something wrong
eulerscheZahl: not t, that whole part
(t*t<a*a||a*a==t*t&t>a)
Maxim251: I know what I did rong
Maxim251: t?:a=0
Maxim251: this is proper
Maxim251: 116
Astrobytes: you get an allowance of 2 English conversations a year in #fr Westicles
Maxim251: But there is or statment
eulerscheZahl: i then butchered the french grammar while writing there and pinged automaton
eulerscheZahl: who responded that he doesn't understand
Westicles: when I visited #fr: Stilgart: merci aux modérateurs de faire le travail et de me dégager le provocateur ci-dessus
eulerscheZahl: that was against you, not against writing English
Astrobytes: I saw that euler :D
Maxim251: Thats so wired,
eulerscheZahl: but when you stay there too long writing english, they want to get rid of you
eulerscheZahl: had a great deepl translation there
Astrobytes: Unless it's of the utmost importance and imperative that it must be dealt with quickly
Maxim251: #eulerscheZahl Yes I found it out how aplay ? : without if statment
Astrobytes: Or when #fr is almost empty
eulerscheZahl: original: CopperFr writes "please write French" (in English)
deepl: fr->en
CopperEng: please write English
Maxim251: >Whole line now is shining at yelow, and still works.. i
Astrobytes: :D
Maxim251: #import<iostream>
int t,a=552;main(){for(std::cin>>t;std::cin>>t;)!(t*t<a*a||a*a==t*t&t>a)?:a=t;t?:a=0;std::cout<<a;}
Maxim251: 117
Maxim251: Hey, earlier was 116
Astrobytes: Good work. Don't keep posting it in case someone copies ;)
jacek: ban
Astrobytes: /ban everything
Maxim251: sorry,
Astrobytes: Hey I don't mind, it's your code
Maxim251: but there must be if statment in one place, now again I get 116
Maxim251: 114
eulerscheZahl: i found it again!
https://imgur.com/a/KSthGGJ
Astrobytes: hahaha
Maxim251: sorry 116 because I try lower value for a, but is not passing validator 2 with big numbers
Maxim251: Then Oficially 116 and my brain want now jump out from my head
Stilgart: Westicles: you are not welcome on #fr, indeed
Westicles: you speak for #fr?
Westicles: self-appointed authority as usual
Maxim251: Thats insane, that ruby can make 49 characters and pas all tests.. There are 1000 peoples with similar resultat, then second place is yavascript 60 characters
Stilgart: self-appointed bastard as usual :)
Westicles: :)
struct: Maxim251 there are shorter than 49 chars
Astrobytes: guys, guys, come on. Not in public eh
MSmits: #catfight
MSmits: I made channel for you
Westicles: We are just joking around, we made up
pb4: https://github.com/marchete/Codingame/blob/master/Optimization/Number%20Shift/NumberShift_LAHC.cpp#L138
MSmits: o
pb4: Anybody understands that nextFloat() function ?
Stilgart: Astrobytes: did the moderators do anything yet ?
Maxim251: #struct Ples dont tell me that stuff... I try kep my brain in place...
pb4: What's the range of the float number that is generated ?
LastRick: Holy moley, while you guys were fighting, I solved Thor 2. What a day, first the stupid cat and now this.
MSmits: do you mean the cat puzzle?
MSmits: or did your cat do something
LastRick: HAH, yes the stupid pixel perfect cat puzzle
MSmits: I'm just asking because euler's cat does stuff
darkhorse64: I have a perl at 48
darkhorse64: oops scroll
MSmits: that sounds like you're an old oyster
LastRick: yes, eulers cat will soon be motioning it is time "fur" sleep (SWIDT)
Astrobytes: MSmits: :rofl:
eulerscheZahl: this morning my cat puked in my room again :/
eulerscheZahl: but not as bad as when she did it inside the open cat door
MSmits: thats what I mean!
eulerscheZahl: so it was inside of my room, outside and in the door itself
MSmits: it's a pro
eulerscheZahl: and i had to open the door and smear it further to get cleaning utensils
MSmits: so evil
MSmits: at least you dont use a roomba
Westicles: pb4, you making a run at NS?
MSmits: maybe he just needs floats
MSmits: we've all been there
pb4: Nope, just randomly saw this code on the forum
MSmits: you randomly saw code that generates random numbers?
DomiKo: is it [0, 1]?
LastRick: and those kind of pet stories are why I have no animals
MSmits: that you know of
LastRick: It doesn't work that way!
MSmits: some rodents come out when you sleep
MSmits: :mouse:
MSmits: euler doesnt have that problem
LastRick: or theres a chipmunk in the parking lot thats really fond of me
MSmits: he only finds pieces of them in a puddle
LastRick: whats worse than finding a dead mouse in your apartment floor?
MSmits: two of them?
LastRick: half of one
eulerscheZahl: a had a night with 2 living mice in my room
eulerscheZahl: cat released them while i was sleeping
LastRick: "released"
Astrobytes: my Gizmo got a sparrow a year or so ago
LastRick: ROFL
MSmits: he caught mice and dropped them in your room euler?
MSmits: is this a reverse cat?
LastRick: hahah
ArianPunk: Guys, can someone tell me how test case number 04 of Dungeons and Maps is not a TRAP?
LastRick: i dont think that cat knows how cats work
MSmits: Gizmo was your cat?
MSmits: Just wondering because thats the name of a gremlin
Astrobytes: Cats aren't overly murderous, they're more sadistic. Once the 'toy' has served it's purpose, they toss it aside
MSmits: dont feed after midnight
LastRick: Arian: You asking about "Many maps" test?
ArianPunk: yes
Astrobytes: That's who she's named after MSmits. The one in my avatar
MSmits: ohh ok
Astrobytes: You can't see her chin here, but it's big like Gizmo
MSmits: so then you know why not to feed after midnight
MSmits: acid reflux right?
Astrobytes: Yes. Also applies to this one.
Astrobytes: hahaha
Astrobytes: pic when she was a kitten:
https://russianaicup.ru/profile/Astrobytes
MSmits: cool
MSmits: she has a nice rating
LastRick: Arian: Haven't done this one in a while but it appears my code finds two solutions and index 2 is the shorter one, so it outputs "2"
Astrobytes: not overly no lol MSmits
ArianPunk: ah
Astrobytes: I actually meant to post the pic link not the whole page :/
ArianPunk: Ok, thnx. I'd think about t
ArianPunk: it
DomiKo: pb4 that union trick is fast division by (1 << 32) - 1
LastRick: If it helps, map 0 also has a solution for that test but its the longer one by two steps. No problem
MSmits: Astrobytes be proud, thats around where Magnus Carlsen is
ArianPunk: i think it's got a problem
ArianPunk: i dont understand why it's not a trap
Astrobytes: MSmits hahaha
ArianPunk: I'm not coding on it. Literally solving it by hand
ArianPunk: There are 5 maps, with 11 by 11 in x and y directions
Astrobytes: arenabot localhost 31002
Astrobytes: woooops
Astrobytes: sorry
MSmits: ArianPunk
MSmits: https://imgflip.com/i/4qho7j
ayoubhs: https://www.codingame.com/clashofcode/clash/150541193338fe5a097d3541accbeb90f1c5773
struct: (╯°□°）╯︵ ┻━┻
jacek: .
Astrobytes: ,
Astrobytes: this is why we need the webchat. Posting punctuation.
jacek: so quiet
Astrobytes: Tends to happen when the chat dies.
MSmits: I can give you a new subject if you want
MSmits: help me improve my breakthrough evaluation function
MSmits: i rank 9, its embarassing :P
MSmits: got a whole bunch of features
MSmits: but doesnt seem to help much
Astrobytes: haven't done it yet
MSmits: ahh ok
MSmits: darkhorse64 had problems with it too
Astrobytes: you're doing an alphabeta
Astrobytes: ?
MSmits: nah same as always, EPT
Astrobytes: ah OK
jacek: you have endtables for breakthrough?
MSmits: i do
jacek: probably bigger and better than mine
MSmits: it's basically a triangular pattern to see if a pawn can race to the end
MSmits: it's checked as soon as a pawn goes on the opposite side of the board
MSmits: so 5th row
Astrobytes: seems reasonable
MSmits: then again when it gets on 6, 7, 8 etc.
MSmits: well 8 is a win
MSmits: but you get the idea
jacek: bunch of features, piece square tables?
MSmits: yeah i have penalty for a piece being on the edge, penalty for empty columns
MSmits: squares that are attacked
MSmits: home guard bonus
MSmits: vertical and horizontal connection bonuses
MSmits: but not bigger patterns
MSmits: it is all local
MSmits: I'd say Ntuple or similar would work here
MSmits: maybe i also just need to prune moves more
Astrobytes: you use ntuples on BT jacek?
jacek: when i had handcrafted eval, i had: piece, square table, if piece is H-connected on row 6+, penalty if piece attacked and not protected (and not my turn)
jacek: no, NN
Astrobytes: ah right
jacek: empty column? did you see that C# breathrough bot?
MSmits: yeah i think so jacek
MSmits: only eval example i could find
MSmits: probably it doesnt work
jacek: it did nothing for me, as well vertical connection. horizontal connection only further in enemy territory
MSmits: I see
Astrobytes: how's your littlegolem bt bot?
MSmits: is this to make sure your pawn doesnt go charging into enemy territory without backup?
Astrobytes: seems like it
jacek: yeah
jacek: though usually against weaker bot single pawn can rush by diagonal unprotected squares
jacek: well, rating 1807, 27 games so far, 27 won
MSmits: yeah but not early, when there's double rows
jacek: though 1 would be lost if opponent didnt make mistake
MSmits: whats littlegolem?
jacek: site with games you can play against players
jacek: correspondence, you get usually 24h for move
jacek: https://www.littlegolem.net/jsp/info/player.jsp?plid=140575
pb4: What is n-tuples ?
MSmits: is it cheating to use an engine?
MSmits: wow pb4 you missed a lot of chat then :)
Astrobytes: Nice. Is that against just bots or mixed? Not sure how they do matchups on there
MSmits: n-tuples is basically looking up a pattern on a board
Astrobytes: No, bots are indicated as such I think MSmits
jacek: they permit it. i bet you read paper about bt bot and they used little golem to see how strong is their bot
MSmits: actually pb4, you sort of did this with yavalath
MSmits: looking up parts of the board for 3 or 4 in a row
MSmits: except instead of looking up the win, you use the result as a key into an array with eval scores
MSmits: and use a bunch of patterns like that to form an evaluation score
pb4: Thanks
pb4: Why that name ?
MSmits: well N because there can be N squares you're looking at
MSmits: and tuple because it consists of multiple parts
Astrobytes: because they are tuples of n-size :D
pb4: Can they be non-neighbours ?
jacek: its just common name. if you look for n-tuples youll find mostly for othello, connect4 and 2048
Astrobytes: Sure, if it fits the game
MSmits: jacek i see, yeah i do remember that paper and the name little golem seemed familiar
jacek: yes
saikoaizen: LO
MSmits: pb4 they can be anywhere on the board
MSmits: but it makes sense to have patterns with squares that affect eachother
pb4: Is there an automatic way to choose which tuples to use, or is it intuition ?
MSmits: well intuition if you dont use a NN
MSmits: NN's kind of naturally makes tuples, except you dont see it
Astrobytes: game-dependent, there might be well-defined patterns
jacek: actually both. you can use GA etc. to create N-tuple architecture and see which is best
Astrobytes: interesting idea
pb4: Gotcha
MSmits: jacek uses them in Othello
MSmits: there's a lot of stuff to google about ntuples in othello
jacek: https://www.researchgate.net/publication/311555004_Systematic_Selection_of_N-Tuple_Networks_for_2048
pb4: Maybe I should try and play one of those board games you've all grinded
MSmits: well you did join on yavalath, but then you stopped
MSmits: you did quite well there
pb4: Yeah, couldn't find a way to improve
MSmits: neither did i, my bot is about as strong as yours still
pb4: wut ? You were first, I was ~5th no ?
MSmits: except of course for the millions of locally simulated games to avoid early game traps...
pb4: ah
MSmits: i mean if i ignore the book :P
pb4: :D
Astrobytes: I was about to comment on that
pb4: I'll soon have a new PC
pb4: muuuuuuch faster
MSmits: good
saikoaizen: Heyo
Astrobytes: so you'll be around here more then?
saikoaizen: New PC dude, that's awesome
MSmits: I only use 1 core each though, for my opening books. It's more about manually adjusting things, letting it run for a while, inspecting etc.
pb4: Don't know Astrobytes, still have to spend a bit of time with the family :o)
Astrobytes: hehe, yeah pb4, always a trade-off :)
MSmits: he needs to build up credit with family, so he can then crush us in a contest
pb4: :innocent:
Astrobytes: hah
pb4: I also have a bandas AI that was 1st when it was still a contribution puzzle
MSmits: ohh nice
MSmits: the top meta uses endgame books now
MSmits: other than that its mostly plain mcts
pb4: Was kinda bummed that the author didn't take any of my comments into account and let an unfinished game be published
MSmits: was one of your comments the fact that the random is unbalanced?
pb4: Don't remember, it was a long time ago :o)
MSmits: ah, well that is the main problem
Astrobytes: I vaguely remember something but yeah
MSmits: I can solve some games at ply 7
MSmits: sometimes its a loss, its a bit frustrating
pb4: https://www.codingame.com/forum/t/community-puzzle-bandas/101048
pb4: Wasn't willing to spend time on a game that was going to change
MSmits: "The game was accepted despite negative pending comments on the random game generation."
pb4: Either it's a contrib and I'm perfectly happy to test it, rewrite my code when breaking changes happen, etc...
pb4: -contrib +WIP
Astrobytes: https://www.codingame.com/contribute/view/21242d55ddfe36f93a624d9d405cc234295e
pb4: or it's been accepted, and they it shouldn't change anymore
MSmits: well if it's an improvement to the random generation, its an ok fix after it goes live, because that shouldnt affect your bot too much
MSmits: but mostly i agree yeah
Astrobytes: (link is original contrib btw)
MSmits: he did improve his random
pb4: iirc map size was changed after the fact
MSmits: increased I guess?
pb4: Yup
MSmits: because if it was smaller I could solve it on ply 1 probably
MSmits: very small branching factor
MSmits: and with the endgame books the game ends at 3x4 size
jacek: map was originally 6x6 but i think he changed it before accepting
jacek: also i think it was before the WIP option
Astrobytes: he did, yes
pb4: You have a bool[500 000] table for the endgame status ?
MSmits: hmm its not that big I think
MSmits: 3^12
pb4: 3^12 ?
pb4: That's 500k :)
jacek: 15
MSmits: oh right
MSmits: i do then
MSmits: it's a string, calculated locally
MSmits: heavily compressed
Astrobytes: physics you say Smits, physics eh
MSmits: :P
Astrobytes: :D
pb4: Compressed how ?
MSmits: symmetry helps a bunch
MSmits: factor of 8
MSmits: 10 results in 1 unicode character
MSmits: so thats a factor of 80 right there
pb4: Nice !
MSmits: i couldnt go beyond the 3x4
MSmits: because there is no 13 sized pattern
MSmits: and the next one would be 14
MSmits: which is 9x bigger in possible combinations
MSmits: also pretty rare to have a 2x7 situation
reCurse: Oh a wild pb4, rare sight
pb4: Hi old guy :innocent:
reCurse: :sweat:
MSmits: lol
Astrobytes: hahaha
MSmits: have you been lurking while we messed with reCurse?
pb4: I might have :sweat_smile:
pb4: I might be lurking far too often
Astrobytes: yes, you should join in more
MSmits: yet ntuples were new to you
MSmits: i bet you dont know jacekmax either
pb4: far too often is not 100% of the time -_-
pb4: Nope, what is it ?
MSmits: jacekmax made a thingy
MSmits: jacek i meand
pb4: @reCurse : new computer incoming, many more cores
reCurse: I feel bullied
pb4: I'll probably give SR another go
MSmits: jacek where's your playground thingy that explains the jacekmax?
Astrobytes: best-first minimax thing guided by UCT
Astrobytes: looking now
Astrobytes: https://www.codingame.com/playgrounds/55004/best-first-minimax-search-with-uct
reCurse: Maybe we should act like civilized people and divide game territory
karliso: I am also puzzled why it is 100ms and not 50ms on bandas.
jacek: theres jacekmax in tags
MSmits: karliso you mean it would save server time ?
MSmits: and have quicker submits?
karliso: No, it would make it less unbalanced.
MSmits: you're right though, the game does not get more interesting for having 100 ms
jacek: bandas is nearly solved
MSmits: ah yes
Astrobytes: karliso, did you give up on raic?
karliso: Astrobytes not at all
karliso: MSmits With 50ms there is a greater chance that a weaker bot will make a mistake.
Astrobytes: oh I saw you were a lot lower earlier, improvement incoming I guess :)
MSmits: yeah i got it karliso, good point
MSmits: I already sometimes see them make a mistake
MSmits: my bot will solve the game as a loss on ply 12 or so, then its a win at ply 18
pb4: Never thought about that "best-first minimax search" before
pb4: I like the idea :)
MSmits: it works fine pb4, i tried it on oware, my bot got to the same rank. I didn't optimize it as well as my other version so it seemed quite a bit weaker still, but maybe with more optim, or different eval it woul dbe just as good as my EPT
pb4: EPT ?
MSmits: mcts with early playout termination
Astrobytes: there may be situations where it's better than ept
MSmits: except my playout terminates at 0
Astrobytes: ^
MSmits: so its basically mcts with an eval instead of a sim
MSmits: its similar to jacekmax, except jacekmax replaces minimax style and my ept version adds up score instead of replacing
pb4: If you could suggest *one* of the board games on which to focus, which one would it be ?
Astrobytes: tough call
MSmits: what do you enjoy about a boardgame
MSmits: do you want one that can be opening booked
MSmits: this could be annoying
jacek: do you want to work on optimizing eval
reCurse: Is there anyone who could possibly enjoy that?
Astrobytes: I can think of *one* person at least
MSmits: :P
MSmits: personally i would recommend Onitama
MSmits: it's pretty to look at and the game has interesting mechanics
MSmits: no opening books
MSmits: also shouldnt take a huge amount of time
MSmits: oh and, minimax and mcts both viable
pb4: Don't really know what to enjoy with board games, I need some kind of intuition of what makes a move good, and I generally don't have that intuition for board games
MSmits: mcts with eval i mean
jacek: or breakthrough. bitboard friendly, many possible features for eval
reCurse: I recommend uttt then
jacek: ill take a screenshot then
MSmits: :grin:
Astrobytes: :D
pb4: I peaked at 80k unguided rollouts on UTTT, with instawin and instaloss-avoidance
MSmits: thats not bad
pb4: Wasn't good enough, and to be fair to reCurse I had absolutely no motivation to write an opening book :D
MSmits: in general i mean
MSmits: also
MSmits: uttt is definitely about more than just sims
jacek: teccles is good opening book
MSmits: theres people in top 10 with 70k
pb4: I seem to remember 100k or 105k to be the reference for the metric I gave
MSmits: yes, but a long time ago either people didnt say anything or everyone just used plain mcts
MSmits: now theres a lot of useful heuristics out that all help a bit. Together they help a lot
MSmits: and opening books are really not that useful unless your opponent is one of the best bots and they play deterministically
pb4: or maybe I should get back to Fantastic Bits :rage:
reCurse: Quality trolling
reCurse: Something I should aspire to
jacek: :no_mouth:
MSmits: what i meant was, in all the games that have a predetermined start, uttt is the game where books are the least useful
MSmits: as far as i can tell
pb4: So many games in the menu, I have trouble finding the ones you suggested -_-
MSmits: really pb4, do onitama
pb4: First time I need ctrl+F on CG...
MSmits: I had the most fun there
reCurse: It's right there in the search bar
MSmits: until i ran out of ideas
struct: :broken_heart:
jacek: try yinsh
pb4: oh wow
MSmits: struct, there is short term fun and long term fun, yavalath was long term fun :P
pb4: Never occurred to me the search bar existed -_-
MSmits: onitama was like a month, i think i spent half a year on yavalath
jacek: and how much on oware?
MSmits: mmh 2-3 months if you dont count the meta mcts stuff
pb4: I still have no idea how you handle transpositions in MCTS on Yavalath
pb4: All the ways I think about seem crazy expensive
MSmits: i dont know if it even works very well, it just helps me solve earlier
MSmits: it is pretty expensive
jacek: he uses it offline in meta mcts
MSmits: halves my rolloutcount
MSmits: or so
jacek: not in actual game
MSmits: no actually i do
jacek: :scream:
MSmits: but it just doesnt help very much
MSmits: its the only game where i do this
MSmits: in meta mcts it helps a ton indeed
MSmits: i like the early solves though
MSmits: thats why i kept it in
jacek: i thought of buying new computer, but there is no available cpu and gpu i want until january/february :rage:
reCurse: Yeah I'm in the same boat
pb4: I'll look up onitama
MSmits: euler made it
MSmits: which is one reason why its good
jacek: he *ported* it
MSmits: good eye for detail
MSmits: sure, the mechanics, but he put it together well with the free art he found
MSmits: it has a good style
MSmits: also they made sure a lot of unbalanced starts where removed. trictrac tried to solve all the possible gamestarts
pb4: gotta go
MSmits: kk nice talking to you
MSmits: gn
pb4: bye
reCurse: bye
pb4: Thanks for the suggestions :)
MSmits: np
struct: bye
Astrobytes: bye!
jacek: c-c-c-combo breaker
Astrobytes: Every. Single. Time.
MSmits: I know you reference killer instinct but i still miss something here
Astrobytes: breaking the chain of similar/same messages/posts
MSmits: ahhhh ok
Astrobytes: it was a forum thing initially iirc. Remember when we all used forums?
jacek: phpbb2?
MSmits: i never used forums that much
MSmits: wait no, memories come back
jacek: ptsd?
MSmits: i used them disguised
MSmits: like, inside a browser game
Astrobytes: lose the 2
jacek: hm?
MSmits: I played astro empires and utopia
MSmits: they had forums
Astrobytes: phpbb
jacek: well it was version
struct: states reduce quite a bit if I take into account that stones are the same color
Astrobytes: I always love your on-topicness struct
Astrobytes: It's a stream of consciousness thing :)
struct: yes, I just say what comes to mind
MSmits: you don't need weed struct
MSmits: you're a natural
Astrobytes: lol
MSmits: lol jacek... I think I just found a line of perfect play for your oware bot, leading to a draw
MSmits: no wonder it took so long
MSmits: well there's no way to counter that
jacek: my bot plays it perfectly?
MSmits: its not a solve, but the entire line of play does not have any alternatives to keep searching. They all lead to the same seed count or worse
MSmits: well i cant be sure
MSmits: but it seems so
MSmits: I found one with re curses bot also
MSmits: he doesnt play it often, but when he does i cant beat him
jacek: in ancient africa you would be oware god
MSmits: thanks, I like that
MSmits: I am assuming here that oware abapa is solved as a draw under CG conditions
MSmits: awari is, but the rules are slightly different and the 200 turn limit might also cause problems
MSmits: I found that if you have 9 seeds left on the board, you can get a different end result depending on how many turns are left, as far back as 145 tirns left
MSmits: so its worse for more seeds
Astrobytes: déjà vu...
Cybersick: Salut le monde
MSmits: exactly, oware has deja vu
MSmits: loops
MSmits: you got it Astrobytes
Cybersick: Je vous invite au nom de l'ENIT Junior Entreprise a notre forum
Cybersick: https://app.fabevent.io/ticket/40?fbclid=IwAR0gK4wVoKsepEiC_OtGmh4QYh1Xqjam7QVmfMIAhMJmCaXGqO7nnpSH7RU
MSmits: hes talking about star trek i think
MSmits: shoo shoo trekkie
struct: #fr
Cybersick: There is an interesting talk about the nasa engineer in a round table
Astrobytes: That's what I get for speaking a french phrase
jacek: oui
karliso: I am curious about oware: if you do openings + NN + endgame tabels code size is supposedly a problem and you have to make a tradeoff between thse 3?
Astrobytes: Cybersick: #fr for French links, also on discord :)
MSmits: I calculate my endgame tables in the first second karliso
MSmits: its not in code side
MSmits: size
MSmits: it is for bandas, but not for oware
MSmits: and I dont use NN
MSmits: thats the rest of the top 5
jacek: all top 4*
Astrobytes: pushed back down?
MSmits: wait what, did they put me at 5 again
jacek: robo added random
Astrobytes: oh, robo submitted earlier I forgot
MSmits: yeah i knew that
MSmits: i thought i went down to 4
jacek: SupSon submitted most recently
MSmits: ah yea i see it
Astrobytes: I've not gone down to 50th or something have I
MSmits: 13
Astrobytes: ah cool, not so bad
MSmits: supson is 0.09 above you
Astrobytes: meh. After raic
struct: Wasnt SupSon top at UTTT?
MSmits: at one time he might have been
MSmits: uttt improved a lot
struct: Yeah but is bot is broken now
struct: I think
struct: his*
SupSon: \ o /
MSmits: how strong was it?
Astrobytes: I may not get a hoodie but I get a t-shirt and I ain't giving up just yet
struct: rank 1
MSmits: yes but compared to which bot?
MSmits: did it beat nagrarok?
struct: yes
MSmits: that bot is unchanged
Astrobytes: good work SupSon
struct: it was around same rank as nagra
MSmits: ahh ok, then supson would now be between 5 and 10
MSmits: People really go wild on uttt, so any top 10 rank is like being nr 1 in a different board game
Astrobytes: We should ping SupSon again
SupSon: what up
MSmits: we're out of subjects to talk about so we picked you
struct: About UTTT
Astrobytes: just thought I'd ping you again since you had so many mentions in the past few minutes :D
struct: your bot is broken for a while
SupSon: yea some compiler change
SupSon: broke me
struct: maybe when the sqrt thing changed?
struct: or something else
MSmits: hard to fix?
Astrobytes: sqrt thing changed?
struct: it was when c++ updated I think
struct: Before you didnt need to include cmath
jacek: he uses complex numbers
MSmits: ohh, yes, be careful about the i
Astrobytes: lol
MSmits: loop using j
MSmits: or k to be safe
Astrobytes: that was ages ago struct
SupSon: does karliso on uttt just have the best opening book
MSmits: no
MSmits: i dont think so
jacek: i bet he uses teccles
Astrobytes: ask him yourself, you pinged him :)
MSmits: when he did have an opening book, i wrecked him 100%, then he turned it off and i had no chance
MSmits: he was strong even without teccles
Astrobytes: I remember that
MSmits: my bot in uttt is 3rd even without book
SupSon: what's teccles
struct: force opponent to play on the board you just played
MSmits: its a heuristic
MSmits: only when the board is empty
MSmits: so always play on the mid edge of mid board and corner of corner board
SupSon: Yes that seems like optimal play for the early game
MSmits: lots of meta mcts confirms this
jacek: my bot does it automatically given few seconds of thinking
MSmits: mine doesnt, what magic is this jacek?
MSmits: you also dont use a lot of sims
Astrobytes: Pony Magic
jacek: yeah, seeing people mentioning 100k on second turn makes me think if i count them properly
SupSon: Is teccles an acronym or something?
MSmits: its a player
Astrobytes: Nah, was (is?) a player
MSmits: i think it was used before him, but he advertised it
jacek: when I have 10k on second turn
MSmits: so how do you count this exactly
MSmits: is this plain mcts you are using?
jacek: 10k, so 10k rollouts
MSmits: and do you playout till end of game?
MSmits: or early termination?
MSmits: with some eval?
jacek: nodes in tree could be greater since i expand all the children, but choose one to do te rollout
jacek: to the end
MSmits: do you use any enhancements?>
MSmits: selection bias?
MSmits: or some kind of preference for moves in the random rollout?
jacek: 1-plu win/lose, selection bias for winning small board or not using move that leads to already ended board
jacek: in rollout moves that win small board have more chance than others
MSmits: what do you mean by 1-ply win/lose?
Astrobytes: check if you win or lose in 1 ply
jacek: yes
MSmits: in the random rollout?
SupSon: he means favor early wins
jacek: rollout and the selection
SupSon: and late losses
jacek: no
MSmits: i do it in the selection, basically that is mcts solver
jacek: a little more intelligent random
MSmits: but not in the rollout
MSmits: does it help a lot to check for wins in the rollout
jacek: well and mcts solver
MSmits: i know others said it works
MSmits: but they say a lot works and then i try it and it does nothing
jacek: i dont remember how much, but id say something like 55%-60%
Astrobytes: ohhh in conjunction with solver
MSmits: i dont understand
MSmits: 55-60% means the winrate for checking win in rollout vs a bot that doesnt, this i got
MSmits: but whats the conjunction with solver
jacek: thats Astrobytes
Astrobytes: this is on top of the fact he has solver implemented
MSmits: what is
jacek: i always have mcts solver
MSmits: me too
MSmits: i just wondered what you meant about 1 ply win/loss in selection
Astrobytes: there
MSmits: i mean doesnt the solver do that?
jacek: depends if you expand only by 1 child instead of all children
MSmits: ohh I see
jacek: and even if by all children, sometimes checking for win/lose is expensive
MSmits: yes when i expand and one of them is a win, the parent is a loss
MSmits: so i stop that iteration
MSmits: is that what you mean?
jacek: exactly
MSmits: ahh ok, to me thats part of normal mcts solver
MSmits: i am confused though
MSmits: two things
MSmits: 1) you're not the best at performance, but usually a lot better than 10x as slow than others
Astrobytes: now that you describe it thus I agree I also thought that was integral to the solver
MSmits: 2) how can you rank this high with 10k
Astrobytes: ^ that's the one
jacek: dunno
Astrobytes: PONIES
MSmits: you literally do the same thing i do except i have a lot more heuristics for pruning and move selection
MSmits: so i shoul dbe slower
jacek: maybe those heuristics suck
MSmits: lol
MSmits: no i tested them of course :P
Astrobytes: haha
MSmits: I have a feeling you hold the key to the rest of us beating kar liso
MSmits: and you dont know what it is
Astrobytes: *karlis o
MSmits: yeah i am being secretive
jacek: karliso?
MSmits: lol
Astrobytes: no, the other one
Cybersick: Thanks for the info Astrobytes :heart_eyes:
karliso: I would like someone to challenge me, but there is no easy way I guess.
MSmits: are you talking about uttt
karliso: yes
MSmits: as you said before, I could just solve your p2 game again
MSmits: and win 100%
karliso: It is not deterministic
MSmits: it is not random either
MSmits: there are only so many lines of play
Astrobytes: de rien Cybersick
MSmits: especially with teccles on
MSmits: you make maybe 10 choices with some choices being fixed, some choices being 2 possibilities
MSmits: so i am guessing at most 200-300 lines of play I need to solve
MSmits: sounds doable
Astrobytes: I sense A Challenge
karliso: I guess you could make unbeatable opening book for p1
MSmits: I can
MSmits: I am messing with you, i already have one that beats you 100% as p2
MSmits: i just havent submitted
MSmits: its not that interesting to me
Astrobytes: Does it beat the others though
MSmits: no
MSmits: I targeted him specifically
MSmits: i can push tomatoes to 1
karliso: Well..
MSmits: with me as 2
karliso: I can change opening move and resubmit
MSmits: yes
MSmits: that is why i did not submit
MSmits: i was just wondering how big i can make it
MSmits: the book
Astrobytes: Do something other than booking?
MSmits: maybe counter most good lines of play
karliso: I don't really care... as I got rid of my opening book.
MSmits: it was the best decision
MSmits: your bot is great and a book would only hurt you
MSmits: i am still wondering what magic must be in those 3k lines of code
MSmits: i am still testing some other ideas Astrobytes, jacek gave me the idea to add random to my own bot so i can selftest
MSmits: i will try out that wincheck
jacek: add random?
MSmits: yeah, so i dont get overtraining issues
MSmits: like, a change only works vs myself
jacek: in uttt?
MSmits: yes, it happens, my bot gets pretty deterministic as well, just like karliso's
MSmits: so random helps
karliso: http://chat.codingame.com/pastebin/646a267c-cbb7-43f4-9d67-6559a5b516ae
karliso: Here is my uttt rollout
MSmits: ohhh
jacek: endgames eh
Astrobytes: <everyone furiously saves to file>
MSmits: my precious...
Astrobytes: (except me, I don't care :D )
MSmits: i dont see endgames?
MSmits: uttt does not seem likely as a game to use endgames with
MSmits: karliso you still get like 400k rollouts with this?
MSmits: it doesnt seem like it would do that
MSmits: ohh i see it
MSmits: A == UTTT::ENDGAME7
MSmits: I'm going to have to stare at this a long time before I figure it out
struct: 400k rollouts?
karliso: I had a bug where I counted early rollouts twice
karliso: So I guess 200-250K turn 2
MSmits: ohh that is much closer to re curse
MSmits:
he had 180k afaik
MSmits: do i read it correctly that you terminate rollouts early?
Astrobytes: that's your kinda bug struct :D
karliso: Not really... unless game is won by 3 in line
MSmits: oh ok
MSmits: did you improve selection somehow, performance wise?
MSmits: I remember you saying once
Astrobytes: just win detection not ept
MSmits: that you felt selection isnt always necessary?
Astrobytes: oh, answered, sorry, don't mind me
karliso: wdym
MSmits: and you left it at that, I kept wondering
MSmits: i dont know, i figured you maybe did random for the first few visits of a parent
MSmits: so you didnt have to do the uct calc
MSmits: or maybe picked with eval
struct: Astrobytes I guess its a normal thing :p
MSmits: the first few times
MSmits: I do think you said you used avx for selection
MSmits: but that did not help me much
karliso: I'm always doing UCT, but that could be improved I guess
MSmits: yeah maybe thats what you meant when you said that, that you thought it could be improved
struct: the problem with avx is that values wont be alligned right?
MSmits: yeah
MSmits: unless you completely mess up your node array
MSmits: never had the courage to try that
struct: But does loading the values take that much time
struct: for it not to be worth?
MSmits: it was about equally fast for me
Astrobytes: for calculating UCT?
MSmits: for doing the sqrts
Astrobytes: ah
MSmits: 8 of them at once
Astrobytes: yeah I know the vectorization bit
MSmits: the problem is that nodes are spaced out 40 bytes each or so
MSmits: so the visitcounts are also
MSmits: at least I think thats the problem
MSmits: you cant just copy a 8 float block
Astrobytes: time to consult reCache
Astrobytes: :)
struct: why 40 bytes?
MSmits: Its just a guess, i dont remember my exact node size
MSmits: it varies per game
MSmits: maybe smaller for uttt
reCurse: What year is this
MSmits: i knew you would say that
Astrobytes: '89
reCurse: It's not conclusive, you could know I would say that yet still be in the past
MSmits: reCurse you'll be relieved to hear that karliso actually had somewhat over 200k rollouts, not 400k
Astrobytes: Or thereabouts. Since Old John submitted his 30 year old Othello code I'm going with that.
MSmits: you can now sleep again
reCurse: Relived? What
MSmits: it was bothering you before
reCurse: Was it
MSmits: oh yeah
MSmits: we both didnt get how it could be so high
reCurse: No such recollection
MSmits: traumatic events tend to do this
reCurse: Maybe a past me travelled to the future then went back
MSmits: sounds right
reCurse: Can't discount possibilities so easily
struct: Its the easiest way to increase sims, I did the same for STC
MSmits: timetravel is the easiest way?
reCurse: With 2020 the way it's been probably
MSmits: how is it for you now?
MSmits: with work and such
reCurse: Work is fine
Astrobytes: Vibrating Thing Theory
MSmits: my job has been somewhat harder with covid
reCurse: Easiest part of the pandemic for me actually
MSmits: I mean i am happy to be at home, but when i was teaching with the mask on it was not fun
Astrobytes: Goobersymmetry, non-Branes and wiggly things that cause trouble
reCurse: Yeah I'd go nuts wearing a mask 8h a day
MSmits: one day, i accidentally brought only one mask, it broke just as i wanted to get into the train
Astrobytes: you use disposable?
MSmits: then i rushed back to get 4 more, just barely got into the next train
MSmits: and 3 more broke during the day
MSmits: usually yes
Astrobytes: jeez, who made those? Us?
MSmits: but thats mostly because my wife accidentally bought a years worth
reCurse: Why disposable? There's plenty of good ones that are reusable
MSmits: yeah
MSmits: I know
reCurse: How do you accidentally...
MSmits: but we have them now =/
MSmits: she didnt do the math
Astrobytes: lol
MSmits: they pack pretty tight
struct: she did ++sims;++sims;
Astrobytes: is this like winning a 10 year supply of toilet paper
MSmits: something like that, except toiletpaper is way bigger
Astrobytes: Well, yeah. That's something at least.
MSmits: anyway the strings are glued and sometimes the glue lets go
Astrobytes: string theory for ya
MSmits: yep :)
MSmits: i probably have to go to work one more time on thursday
MSmits: management decided exam classes still have to be done physically
Astrobytes: gluons turned to gluinos
reCurse: There's a pun that's just too easy there
Astrobytes: Really?
MSmits: ahh yes supersymmetry
Astrobytes: :)
Astrobytes: great subject
MSmits: because of the funny names you mean?
Astrobytes: Why in person?
Astrobytes: No, I find it curious
MSmits: oh, well in person learning is inherently better in most cases
MSmits: and these guys have tests after christmas break
Astrobytes: yes, but pandemic...
MSmits: i know, it will only be one day though, the week after thebreak it will be online
Astrobytes: Fair enough I suppose
MSmits: it's just annoying for me, because on thursday i will have 2 online classes as well
MSmits: also my CS class doesnt have a test
MSmits: so in my case it is not particularly useful
MSmits: also i have an online course
MSmits: so why bother with the in person part
MSmits: but the students are at school all day for other classes....
Astrobytes: Seems wasteful, but we're human right
MSmits: right
SupSon: you guys are humans?
MSmits: to a degree
MSmits: Astrobytes seems the most human
Astrobytes: Why? Cause I have a sore back? :P
MSmits: though when he said he is into supersymmetry for fun, doubts grew
MSmits: because you dont do most of the weird things some of us do
MSmits: it's a compliment :P
Astrobytes: I do RAIC for fks sake
MSmits: thats true
Astrobytes: Isn't that weird enough?
MSmits: yeah its pretty weird
reCurse: That justifies talking about uttt then. Possibly.
Astrobytes: I needed a break
struct: Did they add fog yet?
Astrobytes: Yes, that's round 2
Astrobytes: They just updated with this:
In today’s update, we have fixed a vulnerability in server’s pathfinding: since server is aware of the game state, it could pathfind around obstacles that you do not see in fog of war. Now, server-side pathfinding considers cells out of sight range to be empty.
reCurse: Round 3 has game result decided by observation
Astrobytes: I'm not moaning, I've qualified for a t-shirt
MSmits: thats pretty good right?
MSmits: not sure how strong the opposition is
reCurse: I'm reminded of brood war tricks if my memory serves me right
MSmits: never tried raic
struct: I was thinking on doing
struct: then I looked at the game
reCurse: You could figure out where the guy is located if he decided to wall his entrance
reCurse: By looking at where your unit would go by moving there
MSmits: ohh
MSmits: smart
Astrobytes: heh heh, yes
MSmits: so you leave a little hole in the wall to counter this?
reCurse: There's a good reason you're walling
reCurse: To prevent run-ins
reCurse: So no, you don't
MSmits: well yes but its poor formation if they have to go through a small crack
Astrobytes: but it's really not SC this game
MSmits: maybe a small hole with some turrets
MSmits: tower defense :)
Astrobytes: And I'm not the best person to ask about the real meta rn, since I'm around 179th
reCurse: What is the real meta Astrobytes
MSmits: we dont need the best person
Astrobytes: Turret rush!
reCurse: So how is that not SC again?
Astrobytes: It's not the meta I only said that to bait you
reCurse: Real bunch of meanies today in the chat
reCurse: Must be MSmits presence
Astrobytes: :D
Astrobytes: speaking of which, nice to have you back MSmits
Astrobytes: in chat
MSmits: thanks Astrobytes, feels good to be back
MSmits: might still not be as active as i want to be. Getting this degree takes more time than i expected
MSmits: I got about 6hrs a week work relief to do it, but it takes more than 6 hrs a week
karliso: https://github.com/karlisolte/ultimate-tic-tac-toe
karliso: Sharpen your uttt tactis here
Astrobytes: seriously though, meta seems to be ranged units only, just one barracks (which I was surprised about and am exploring), careful supply depot placement and targeting enemies well
Astrobytes: haha. The man.
Astrobytes: Anything to add to that RAIC meta karliso?
karliso: I thought walls are useless
MSmits: lol karliso, i like it, but what is your point with this github?
DomiKo: How that happened that we have learned that much about UTTT today
Astrobytes: Thought? You have started to use them?
Astrobytes: Walls and turrets seemed useless tbh
karliso: I collected pictures of "tactics" and made probably first uttt tactics quiz.
Astrobytes: The former more than the latter
MSmits: ohh, it reads a lot like a chess puzzle indeed
MSmits: I might do something like this for yavalath
MSmits: for my study project
Astrobytes: 5D Super Lavayath
MSmits: with the volcano!
Astrobytes: Supersymmetry! Now that would add something
Astrobytes: Scratch that. My brain started overheating at the possibilities
MSmits: karliso i was worried you shared your bot when you posted that gith
Astrobytes: Have more faith
MSmits: earlier he shared his rollout
MSmits: thats why :P
Astrobytes: Worthless without context and understanding
MSmits: if you reverse more words, you'll go all yoda
Astrobytes: When did I reverse
MSmits: you didnt, i am saying thats all that's missing
Astrobytes: Worthless without context and understanding this is, hmm
MSmits: worthless without context, this is
MSmits: exactly :P
Astrobytes: sorry, it's late :D
MSmits: yeah, i still dont get why that guy thought our late chat is less fun
struct: Usually its not this active
Astrobytes: Nah, he's cool, he meant the non-european hours
struct: At around this time
MSmits: isnt this non european?
struct: Give it 20 more minutes and you will see lag ocmplaints
MSmits: ahh ok
struct: complaints*
MSmits: I guess i always miss this
MSmits: because of sleep
Astrobytes: after this point, unless a bunch of us are active it's full of idiots, complaints and trolls
MSmits: I see
struct: Did not do much progress on STC today
Astrobytes: yeah, the lag from the updates etc
DomiKo: struct you doing STC now?
struct: Im storing states until depth N like euler suggested
MSmits: you're really into STC struct
struct: DomiKo I have been doing it for a bit
struct: then paused
DomiKo: I see
struct: I wasted 1 week on trying to make a fast sim
struct: not worth
Astrobytes: don't do the bitboard thing then
struct: I wont
DomiKo: I wasted 5H trying to beat boss gold with wrong formula for score...
struct: My first sim had like 30k sims on c++
DomiKo: sim is 8steps?
Astrobytes: I could have sworn you had a legend bot before
struct: no 1 depth
struct: I had
struct: my bot is like 2years old
struct: Then I worked on optimization and never submited
struct: Now I decided to rewrite
Astrobytes: that was it
Astrobytes: well, anyway, I'm all done for tonight
struct: gn
Astrobytes: see you all later
DomiKo: See you
MSmits: gn!
MSmits: going also
struct: gn
FrankGotzTank: Why does my program say that 1000 / 60 = 16.666666666666668???
struct: precision
struct: its normal
FrankGotzTank: But it should be 16.66666666667
struct: Which language are you using?
FrankGotzTank: C#
struct: hmm
struct: it gives me right value
struct: https://tech.io/snippet/oAOhQ3t
struct: ah on CG its different
FrankGotzTank: yeah right
FrankGotzTank: That's pretty annoying
Kamikaze: does anyone can help me with ultimate tic tac toe? i was trying to learn the minimax algorithm but i dont know what to do, im trying to make it with python3
VizGhar: uf... you can't solve UTTT solely with minimax
VizGhar: if you just want to try minimax, try it on less complex problem (clasic tic tac toe)
LightFury01: hey guys
