pranshu-raj-211/project-alphazero

Chess engine in Python, currently in development

Hi there 👋

I'm Pranshu Raj, a sophomore at IIIT Gwalior pursuing B.Tech in Computer Science.

• 🔭 I’m currently working on a Reinforcement Learning agent for playing games
• 🌱 I’m currently learning MLOPS and improving my python knowledge
• 📫 How to reach me: linkedin [email protected]

project-alphazero's Issues

add validation for moves in game tree

Tests for verifying correctness for movegen, verify against stockfish results.

Fix eval in case of checkmates

Currently my minimax function only checks if the depth is zero or the game is over, and returns the eval.
This prevents it from detecting checkmates and responding correctly, which leads to it missing mates, even mate in 1.

Add pgn saving feature

Need to save pgn of games which have been played to allow verification of moves

replace pd dummies by sklearn OHE

Add some variety to the openings

Maybe use a random number generator at first, but add opening books (atleast a couple openings).

Prevent three fold repetition

documentation

Prepare documentation on the code as well as the process

How should I do the outlier removal?

Outlier removal is done here using quartile method but there are still many issues with it, such as the size/bhk value should be less than 300 but there are still many values outside of that and dropping them or imputing may affect the results drastically.

New features to add

Categorical encoding
Stratified sampling (to avoid imbalances) or weighted sampling
pipelines for the processing, or use classes
Add more visuals

implement puzzle playing

Implement lru cache

Save some board positions for things like iterative deepening in ab search - save the moves used to prune further.
Can also use this for the history heuristic.

Where should I drop the duplicated rows in the dataset

The dataset has around 900 duplicated rows which I want to drop but I cannot implement it without causing other things to break and I'm also confused about where to drop the duplicates exactly.

automate workflows

need to add pipelines to automate the preprocessing part

Add quiescence search for capture moves

Move ordering - checks, captures, attacks and so on

implement a way to sort all the legal moves in the given order for improving minimax efficiency

Add dataclassses for cleaner state representation