ChessWinnerPrediction

---

Content

• Project Description
• Installation and Running a Demo
• Data
• Project Organization
• Static Move Solution
  • Dataset Overview
  • Hist Gradient Boosting Classifier

Project Description

The Chess Winner Prediction project is a machine learning initiative following the Cookiecutter Data Science structure. The goal is to predict the winner of a chess game using two distinct approaches:

Solution	Description
Baseline Solution	Focuses on pre-game information, such as player Elo ratings and game base time. This straightforward model acts as the foundation for comparison.
Static Move Solution	Utilizes in-game data, including move evaluations, remaining time, and the number of pieces on the board, to enhance prediction accuracy.

Installation and Running a Demo

To get started with the Chess Winner Prediction project, follow the steps below:

Clone the Repository

git clone https://github.com/Tikhon-Radkevich/Chess-Winner-Prediction.git
cd Chess-Winner-Prediction

Set Up the Virtual Environment

python -m venv .venv
source .venv/bin/activate

Install Dependencies

pip install -r requirements.txt
pip install -e .

Run the Demo

streamlit run ./demo/main.py

---

---

Data

Dataset source: lichess.org standard games
Dataset represented as .pgn.zst archive files.

Sample Example:

[Event "Rated Bullet tournament]      [Site "https://lichess.org/PpwPOZMq"]
[White "Abbot"]                       [Black "Costello"]
[WhiteElo "2100"]                     [BlackElo "2000"]
[WhiteRatingDiff "-4"]                [BlackRatingDiff "+1"]
[Result "0-1"]                        [ECO "B30"]
[TimeControl "300+0"]                 [Termination "Time forfeit"]

1. e4 { [%eval 0.17] [%clk 0:00:30] } 1... c5 { [%eval 0.19] [%clk 0:00:30] }
2. Nf3 { [%eval 0.25] [%clk 0:00:13] } 2... Nc6 { [%eval 0.33] [%clk 0:00:27] }
3. b3?? { [%eval -4.14] [%clk 0:00:02] } 3... Nf4? { [%eval -2.73] [%clk 0:00:21] } 0-1

Project Organization

project structure based on Cookiecutter Data Science

├── chesswinnerprediction        <- Source code for use in this project.
│   │
│   ├── baseline           <- source code for baseline solution: models, data processing and visualization utils.
│   ├── static_move        <- source code for static move solution.
│   ├── dataloader         <- Scripts to download and process zst archive from lichess.org to csv.
│   ├── processing         <- Scripts to preprocess data for baseline and static move solutions.
│   └── visualizations     <- Functions to create exploratory and results oriented visualizations
│
├── notebooks                    <- Jupyter notebooks: models, visualizations, and data exploration.
│   │
│   ├── baseline_notebooks <- baseline solution: reamde file, notebooks with different models like: 
│   │                          logistic regression, knn, desicion tree, random forest and boosting.
│   └── static_move        <- Solution based on Hist Gradient Boosting Classifier from scikit-learn.
│
├── scripts                      <- Scripts to download and process zst archive from lichess.org to csv. 
│                                   Also, scripts to preprocess data for baseline and static move solutions.
│
├── data                         <- Data files for use in this project, including example data for demo.
│
├── models                       <- Trained baseline and static move models for demo.
│
├── demo                         <- Streamlit web app for demo.
│
├── README.md          <- The top-level README for developers using this project.
│
├── requirements.in    <- The high-level requirements.
│
└── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
                          generated with `pip-tools compile -o requirements.txt`

---

Static Move Solution

Dataset Overview

Dataset source: lichess_db_standard_rated_2017-05

Result	Event	eval	EloDiff	MeanElo	BaseTime	IncrementTime	time_diff_norm	white_remaining_time_norm	black_remaining_time_norm	white_time_per_move	black_time_per_move	white_increment_pct_in_time_per_move	black_increment_pct_in_time_per_move	white_time_will_end_on_move	black_time_will_end_on_move	i_move	w_score	b_score	n_pieces
0-1	Rated Blitz game	6.72	-41	1580.5	300	0	-0.323333	0.256667	0.580000	0.016895	0.009546	0.0	0.0	15.191926	60.75554	44	35	34	27
1-0	Rated Blitz game	0.10	18	1572.0	300	0	-0.013333	0.910000	0.923333	0.003914	0.003334	0.0	0.0	200.000000	200.00000	23	38	38	30

• Result: The target variable representing the outcome of the game.
• i_move: The number of moves played so far, including both white and black moves.
• EloDiff: The difference in Elo ratings between White and Black players, calculated as WhiteElo - BlackElo.
• remaining_time_norm: The remaining time for a player normalized by the base time, where the lower bound is 0.0 (no time left) and the upper bound is 1.0 (full base time remaining).
• time_diff_norm: The difference in normalized remaining time between white and black, calculated as white_remaining_time_norm - black_remaining_time_norm.
• time_per_move: The average time per move, calculated as (1 - time_diff_norm) / i_move.
• time_will_end_on_move: The projected move number when the player will run out of time, calculated as remaining_time_norm / time_per_move, where the upper bound is 200.
• increment_pct_in_time_per_move: The percentage of the increment time per move, calculated as increment_time / time_per_move.
• score: The material score, where each piece is assigned a value:
  • 1 for a pawn
  • 3 for a knight or bishop
  • 5 for a rook
  • 9 for a queen

Target to i_move Distribution:

Hist Gradient Boosting Classifier

Notebook: histGBC_implementation.ipynb