Oden Chess Dataset

Dataset Summary

The Oden Chess Dataset is a comprehensive collection of over 4 million chess games compiled from top players, major tournaments, and categorized by opening systems. This dataset provides rich annotations including move sequences, board positions, player information, and game metadata, making it ideal for chess AI research, opening analysis, and statistical studies.

Dataset Details

Total Games: 4,047,908
Source Files: 299 PGN files
Total Size: ~2.6 GB (original PGN format)
Total SizeHF: 10.1 GB
Time Period: Historical games to 2024
Languages: English (PGN notation is universal)

Dataset Structure

Data Fields

Each game in the dataset contains the following fields:

Field	Type	Description
`event`	string	Tournament or match name
`site`	string	Location where the game was played
`date`	string	Date in YYYY.MM.DD format
`round`	string	Round number in tournament
`white`	string	Name of player with white pieces
`black`	string	Name of player with black pieces
`result`	string	Game result: "1-0" (white wins), "0-1" (black wins), "1/2-1/2" (draw), "*" (unfinished)
`white_elo`	string	White player's ELO rating
`black_elo`	string	Black player's ELO rating
`eco`	string	Encyclopedia of Chess Openings (ECO) code
`opening`	string	Opening name
`variation`	string	Opening variation
`white_title`	string	White player's title (GM, IM, FM, etc.)
`black_title`	string	Black player's title
`time_control`	string	Time control format
`termination`	string	How the game ended
`moves`	list[string]	List of moves in Standard Algebraic Notation (SAN)
`moves_san`	string	All moves as a single string
`positions_fen`	list[string]	Board position in FEN notation after each move
`num_moves`	int32	Total number of moves in the game
`tags`	list[string]	Categorical tags based on source
`source_file`	string	Original PGN filename
`all_headers`	string	JSON string of all PGN headers

Data Organization

The dataset is organized into three main categories:

Players: Games from individual top players including world champions and grandmasters
Openings: Games categorized by opening systems:
- Classical King Pawn (e.g., Italian Game, Spanish Opening)
- Classical Queen Pawn (e.g., Queen's Gambit)
- Modern King Pawn (e.g., Sicilian Defense, French Defense)
- Modern Queen Pawn (e.g., King's Indian, Nimzo-Indian)
- Flank and Unorthodox (e.g., English Opening, Bird's Opening)
Tournaments: Games from major championships including World Championships, Candidates tournaments, and other elite events

Usage Examples

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("BBSRguy/Oden-worldchess")

# Access the training split
chess_games = dataset['train']

# View a sample game
sample_game = chess_games[0]
print(f"White: {sample_game['white']} ({sample_game['white_elo']})")
print(f"Black: {sample_game['black']} ({sample_game['black_elo']})")
print(f"Result: {sample_game['result']}")
print(f"Opening: {sample_game['eco']} - {sample_game['opening']}")
print(f"Moves: {sample_game['moves_san'][:50]}...")

Filtering Games

# Filter games by player
carlsen_games = chess_games.filter(
    lambda x: 'Carlsen' in x['white'] or 'Carlsen' in x['black']
)

# Filter games by opening
sicilian_games = chess_games.filter(
    lambda x: x['eco'].startswith('B') if x['eco'] else False
)

# Filter games by result
decisive_games = chess_games.filter(
    lambda x: x['result'] in ['1-0', '0-1']
)

# Filter long games
long_games = chess_games.filter(
    lambda x: x['num_moves'] > 100
)

Analyzing Positions

import chess
import chess.svg

# Reconstruct a game position
game = chess_games[0]
board = chess.Board()

# Play through the moves
for move in game['moves']:
    board.push_san(move)

# Or directly load a position
position_after_10_moves = game['positions_fen'][10]
board = chess.Board(position_after_10_moves)

Statistical Analysis

import pandas as pd

# Convert to pandas for analysis
df = chess_games.to_pandas()

# Result distribution
print(df['result'].value_counts())

# Most common openings
print(df['eco'].value_counts().head(10))

# Average game length by result
print(df.groupby('result')['num_moves'].mean())

# Top players by number of games
all_players = pd.concat([df['white'], df['black']])
print(all_players.value_counts().head(20))

Applications

This dataset is suitable for:

Chess Engine Development
- Training neural networks for position evaluation
- Move prediction and game analysis
- Opening book generation
Statistical Analysis
- Player performance metrics
- Opening popularity and success rates
- Game length patterns
- ELO rating analysis
Machine Learning Research
- Sequence modeling with chess moves
- Pattern recognition in positions
- Reinforcement learning for chess AI
Educational Tools
- Opening repertoire builders
- Tactical pattern recognition
- Historical game analysis

Dataset Creation

Source Data

The dataset was compiled from publicly available PGN files including:

Individual collections of top-rated players
Major tournament archives
Opening-specific game collections

Processing

Games were processed to:

Extract all metadata from PGN headers
Convert moves to a standardized format
Generate FEN positions for each move
Categorize games by source type
Handle various PGN format variations

Considerations

Data Quality

Some games may have incomplete metadata (missing ELO ratings, dates, etc.)
A small number of games contain annotation errors from source files
Time controls and termination reasons may not be available for older games

Ethical Considerations

All games are from public sources and chess games are not copyrightable
Player names are included as they appear in public tournament records
No private or sensitive information is included

Citation

If you use this dataset in your research, please cite:

@dataset{oden-worldchess,
  title = {Oden World Chess Dataset},
  author = {BBSRguy},
  year = {2025},
  month = {6},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/datasets/BBSRguy/Oden-worldchess}},
  note = {A comprehensive chess dataset with 4M+ games from top players and tournaments}
}

License

This dataset is released under the MIT License. Chess games themselves are factual records of public events and are not subject to copyright.

Acknowledgments

Thanks to all the chess organizations, players, and enthusiasts who have made these games publicly available for analysis and study. 😊 🙏

Datasets:

BBSRguy
/

Oden-worldchess

You need to agree to share your contact information to access this dataset