v1.0.0

Goal

The primary objective of this release is to enhance the chess engine's ability to predict the best move from a list of possible moves, as determined by the choice an expert player would make. The model aims to rank moves by learning a relative urgency metric that reflects the likelihood of an expert choosing a particular move over others.

Modeling Approach

The model is trained using a dataset comprising 20,000 games played by players with an ELO rating of 2400+. The underlying principle is that the move selected by an expert among all possible moves has the highest urgency value. The urgency represents the priority or importance of a move as judged by its expected effectiveness in the game.

To evaluate moves, the model plays each possible move and ranks the resulting board states against eachother. Directly storing and comparing all possible board values/urgencies is infeasible due to the astronomical number of possible configurations (>10^80). Therefore, we decompose the board into a set of features and represent the board state as the sum of these feature values/urgencies. Currently, the possible moves from a board represent its features. This results in a maximum of 64 x 64 x 12 features.

Model

Files

• training_data

• trained_model

Accuracy