Using Machine Learning to play chess

Aubhro Sengupta
pyohio.org/sche...
Today, chess engines contain many lines of code handcrafted under the guidance of grandmasters. Are you interested in being knee deep in chess theory just to crank out a half decent engine? No? This talk is for you. Why not create an engine that learns to improve itself?
PyOhio is a free (thanks sponsors!) annual conference for Python programmers in and around Ohio and the entire Midwest.
pyohio.org/

[@stevenmcgown6552]

To fix the problem with openings, you should have it randomly choose from textbook moves instead of just randomly moving the pawns.

[@Lord_Liverpool]

Outstanding! Well done young man!!

[@MrJerryHsu]

Our boys Eric and Aman @ 20:00 xD

[@AnupKumar-wk8ed]

Very nice work!

[@maheshsethi8593]

Is it possible to predict the moves on the first move to win throughout the game, in short prediction of moves by choosing out the probabilities of opponents moves

[@TrustPatches]

Look into the Minimax algorithm, you could look at maybe using a neural net to produce heuristic values for that algorithm.

[@glitched_code]

Chess engines already do this, but the result may vary in a stochastically random way, since the starting position along with a fair number of opening lines are considered to be drawn. Probabilities of a certain move happening become irrelevant if we consider the engine to always play the best move on both sides.

[@chrisdavis2161]

in short, no (because of the nature of the game) with "perfect" moves on both sides from the starting position the game draws. Chess engines try to predict the best moves based on their data sets. I don't believe that probability matters too much in their calculations like in human chess (you can learn traps based on probability) but engines nowadays don't fall for human traps. The traps they use are so far above human knowledge that it would be impossible to recognize at first glance. I use "perfect" because the engine doesn't always play the objective best move. It plays what it thinks is the best move at the time, until it is proven wrong by a better set of move ideas. The idea that leads to a win was always the "best" move based on the playing strength of both players

[@jamesbowman1948]

Awesome

[@chesssuite3243]

Leela Zero open source project solved all the work

[@joaoguazzelli3236]

can i make it during my vacation?? kkkkkk (only two weeks)

[@sebastianjost]

You can try. Better start and finish it later than not trying it at all

[@rinielvillacater1321]

@@sebastianjostlittle too late buddy

[@sebastianjost]

@@rinielvillacater1321 there are people reading the comments later and may have the same question. So why not leave an answer for those?

[@glitched_code]

can i make this in a week?

[@joekanaan2548]

@@glitched_code You can try. Better start and finish it later than not trying it at all

[@GPCTM]

4:37 The "assuming my opponent plays well" is my weapon to, sometimes, beat computers. I have done it and I don't play chess. :-) It's easier beating level 9 than level 1 because of that (me not knowing chess and the computer assuming I am a genius).

[@Rocketos]

If you are not lying , I would love to see you beating computer level 9 !!

[@GPCTM]

I have only found a screen for level 6 (the first time it happened I was dumbfounded and even wrote that in my blog) acercadomundo.blogspot.com/search?q=xadrez but I know I have done it more times I can't swear on my life about the exact level but I don't think that's important because, remember, I couldn't beat level 1...! and in a few quick tries I was victorious in the highest levels. In high levels, ultra fast playing - as in "I do not know what I'm doing"! - it's possible to win! This was on a windows Vista that no longer exists. I recall that the only problem for higher levels was the unbearable long thinking time... But I know I have done better than 6 and had make more screenshots. If I discover more I will show. You try it: fast, furious and fearless play will eventually make you victorious, I'm sure.

[@Rocketos]

Yea, my chess elo today is 1600, so I know more than just a little about the game. But you said you don't know very well to play it, so it was impressive that you won agains level 9 computer (assuming you didn't use UNDO moves) because the software is programmed not only with the "thinking" of doing things considering you are a high level player, it calculates the moves seeking for optimal values of a function previous defined...

[@sl1y68]

a bishop is not worth 3.5

[@_productivity__nill_1131]

Sl1Y thought they were equivalent to knights but it varies in an open and closed game.

[@sl1y68]

this true, but according to Fischer and Garry Kasparov they are not worth more than 3.15 pawns

[@amun1040]

@MorTobXD also, there are compounded value like having the bishop pair is more valuable than having a knight and a bishop.

[@hitechj5057]

Practically it’s not, however in some instances it’s actually more valuable than a Knight

[@hydrarl3869]

@@sl1y68 Pretty sure Fischer said 3.25 and Kasparov said 3.15

[@WahranRai]

What about preparing and testing your presentation before ?

[@geetaagarwal4622]

Lol