ArkadiumInc/ArkadiumBackgammon

Arkadium Backgammon Dataset High-Quality Human Backgammon Gameplay (Logs + Optional Board Screenshots)

Opening Position

Mid-Game Strategy

Bear-Off Phase

Dataset Overview

This dataset contains real human gameplay from Arkadium's backgammon platform, providing researchers and developers with high-quality training data for game AI, reinforcement learning, and computer vision applications.

RAW data only. The primary dataset is the raw ZIP archives (.mat/.webp). Parquet files in data/ contain only 10 example records for the HuggingFace Dataset Viewer and are not a full export. We do not provide a full Parquet version.

Totals (both ZIPs): 10,000 matches, ~3.92 GB raw ZIP size.

Metric	gamelogs	gamelogs_with_images
Raw match logs	8,472	1,528
Raw screenshots	-	81,420
Raw ZIP size	~9.7 MB	~3.9 GB
Raw format	MAT (ZIP)	MAT + WebP (ZIP)

Use Cases

• Game AI Training: Train neural networks to predict optimal moves using supervised learning on human gameplay.
• Reinforcement Learning: Use as demonstration data for imitation learning or to initialize RL agents.
• Computer Vision: Board state recognition, piece detection, and game state extraction from images.
• Strategy Analysis: Study human decision-making patterns, opening strategies, and endgame techniques.

Data Format

File Structure

Arkadium Backgammon Dataset
│
├── Arkadium_Backgammon_full_data_gamelogs_001.zip
│   │   └── 8,472 match logs
│   ├── 1200001.mat
│   ├── 1200002.mat
│   └── ...
│
└── Arkadium_Backgammon_full_data_gamelogs_and_images_002.zip
    │   └── 1,528 matches with screenshots
    ├── 0217550.mat
    ├── attachments_0217550/
    │   ├── 0217550_0001.webp
    │   ├── 0217550_0002.webp
    │   └── ... (20-176 screenshots per match)
    └── ...

Game Log Format (.mat)

Each .mat file contains a complete match record in a human-readable format:

; [Place "Arkadium"]
; [Player 1 "pl1"]
; [Player 2 "pl2"]
; [EventDate "2025.06.05"]
; [EventTime "15.46"]
; [CubeLimit "64"]
; [GP "987b847a-9ed5-9e0e-1ef4-b46757caeb39"]
; [RE "pl1+6"]

 3 point match

 Game 1
 pl1 : 0                  pl2 : 0
  1) 35: 8/3 6/3                 43: 24/20 24/21
  2) 61: 13/7 7/6                26: 21/15 15/13
  3) 33: 8/5* 8/5 24/21 24/21    54: 25/21 13/8
      ...
      Wins 6 points and the match

Header Field Reference

Field	Description	Example
[Place]	Gaming platform	"Arkadium"
[Player 1]	First player ID	"pl1"
[Player 2]	Second player ID	"pl2"
[EventDate]	Match date (YYYY.MM.DD)	"2025.06.05"
[EventTime]	Match time (HH.MM)	"15.46"
[CubeLimit]	Maximum doubling cube value	"64"
[GP]	Unique game UUID	"987b847a-..."
[RE]	Match result (winner + points)	"pl1+6"

Move Notation Guide

Notation	Meaning	Example
X/Y	Move checker from point X to point Y	24/20
X/Y*	Move and hit opponent's blot	8/5*
X/0	Bear off checker from point X	6/0
Doubles => N	Offer doubling cube at value N	Doubles => 2
Takes	Accept doubled stakes	—
Drops	Decline and forfeit game	—

---

Parquet Preview (10 records)

Parquet files in `data/` are **only a 10-record preview** for the HuggingFace Dataset Viewer. They exist **exclusively for demo/browsing** and **do not represent the full dataset**. All primary data is in the raw ZIP archives above, and no full Parquet release is provided.

Quick Start (Preview)

Installation

pip install datasets pillow

Load the Preview

from datasets import load_dataset

# Load game logs preview (10 records)
ds = load_dataset("ArkadiumInc/ArkadiumBackgammon", "gamelogs")

# Load with board screenshots preview (10 records)
ds_img = load_dataset("ArkadiumInc/ArkadiumBackgammon", "gamelogs_with_images")

Basic Usage

# Access a single match
match = ds["train"][0]

print(f"Players: {match['player1']} vs {match['player2']}")
print(f"Date: {match['event_date']}")
print(f"Result: {match['result']}")
print(f"\nMoves:\n{match['moves']}")

Working with Screenshots

from PIL import Image
import io

# Get a match with images
match = ds_img["train"][0]

print(f"This match has {match['num_images']} board screenshots")

# Display the first board state
if match['images']:
    img = Image.open(io.BytesIO(match['images'][0]))
    img.show()

Parquet Schema (Preview)

These columns describe the Parquet preview (10 records).

Column	Type	Description	Config
data_id	string	Unique record identifier	Both
file_id	string	Source file reference	Both
place	string	Gaming platform	Both
num_images	int32	Screenshot count	Images only
images	list[Image]	WebP board screenshots (stored as bytes/path in Parquet preview)	Images only
player1	string	First player ID	Both
player2	string	Second player ID	Both
event_date	string	Match date (YYYY.MM.DD)	Both
event_time	string	Match time (HH.MM)	Both
cube_limit	int32	Maximum cube value	Both
match_length	int32	Points needed to win	Both
game_uuid	string	Unique game identifier	Both
result	string	Match outcome	Both
moves	string	Parsed move sequence	Both
raw_log	string	Original .mat content	Both

Citation

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

@dataset{arkadium_backgammon_2026,
  title     = {Arkadium Backgammon Dataset: Human Gameplay with Board Screenshots},
  author    = {Arkadium Inc.},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/ArkadiumInc/ArkadiumBackgammon}
}

Report Issues · Request Features