This document describes a Python program for a chess AI that utilizes a Convolutional Neural Network (CNN) for position evaluation combined with a traditional piece value, alpha-beta search, and position scoring approach.

Project Structure: preprocessed_data.zip has all the preprocessing scripts and datasets we used to train the model gameState.py: This file defines a GameState class representing the chessboard state, including board representation, piece locations, move generation, and evaluation functions. model.py: This file handles training and deploying the CNN model for chess position evaluation. It reads FEN strings representing chess positions, trains the model on a dataset with material-based evaluation labels, and saves the trained model for later use. find_move.py: This file contains the core logic for finding the best move for the AI. It uses NegaMax pruning with alpha-beta search for efficient move exploration. The evaluation function combines the CNN prediction with a traditional piece value and position scoring approach. chessmain.py: Chess logic and board representation which It manages with the chess game logic, keeps track of the board state, and validates legal moves. Graphical User Interface (GUI): It creates the chessboard and piece graphics, handles user input (mouse clicks), and displays move animations and game information. AI Integration : It allows the computer to play as black using a separate process to find the best move and make decisions. cnn_scorer.py: FEN to board matrix: It converts a standard chess notation (FEN) into a numerical board representation for the CNN model.

Key Functionalities:

Board Representation: The chessboard is represented as a 2D numpy array with piece codes (e.g., "wP" for white pawn, "bR" for black rook).

Move Generation: The GameState class provides functions for generating valid moves for the current player considering piece types, movement rules, and checks.

CNN Model: The model.py script trains a CNN model on FEN strings and corresponding material-based evaluation labels. The model predicts a score for a given chess position.

Evaluation Function: The scoreBoard function combines the CNN prediction with a traditional piece value and position scoring approach for a more comprehensive evaluation.

Move Search: The findBestMove function utilizes NegaMax pruning with alpha-beta search to explore possible moves and find the one leading to the best evaluation score for the AI player's color.

Prerequisites: Python 3 with necessary libraries (numpy, tensorflow, scikit learn and pygame etc.) installed. A pre-trained chess_cnn_model.h5 file.