---
license: mit
task_categories:
- reinforcement-learning
language:
- en
tags:
- hive
- boardgame
pretty_name: Hive Match History Dataset
size_categories:
- 100K<n<1M
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: uhp_string
    dtype: string
  - name: player_white
    dtype: string
  - name: player_black
    dtype: string
  - name: winner
    dtype: string
  splits:
  - name: train
    num_bytes: 57507362
    num_examples: 140594
  download_size: 26416714
  dataset_size: 57507362
---

## Dataset Summary

This dataset contains recorded match histories for the board game **Hive**, collected from the online platforms **BoardGameArena** and **BoardSpace**. All matches were played between human players; no AI or bot games are included.

Each record represents a complete Hive game encoded in the **[Universal Hive Protocol (UHP)](https://github.com/jonthysell/Mzinga/wiki/UniversalHiveProtocol)** GameString format, describing every move taken throughout the match.

Games include both base Hive and variations with different combinations of expansion pieces (Mosquito, Ladybug, Pillbug). All entries correspond to finished games that ended with a valid outcome.

~140,000 match histories.

---

### Data Curation

Games were sourced from:
- [BoardGameArena](https://boardgamearena.com)
- [BoardSpace](https://www.boardspace.net)

Filtering criteria:
- Only fully completed games were included.
- Games that were truncated, invalid, or ended in early surrender were removed.
- Remaining games end in one of:
  - White win
  - Black win
  - Draw (by in-game rules or mutual agreement)

The dataset primarily consists of games played with all three expansion pieces (Base+MLP) but includes matches with any subset of expansions as well.

Note that some player names will appear null or empty as their profile data has since been removed.

---

### Data Fields

Each row in the dataset represents one completed Hive game. This does not guarantee that every state within each game is unique since every move is recorded.

| Field | Type | Description |
|--------|------|--------------|
| `uhp_string` | `string` | Complete match record in Universal Hive Protocol (UHP) format. |
| `white_player` | `string` | Name or identifier of the White player. |
| `black_player` | `string` | Name or identifier of the Black player. |
| `winner` | `string` | Name or identifier of the winner, or `"draw"`. |

---

### Potential Uses

This dataset is could be used for:
- **Reinforcement learning** — training and evaluating Hive-playing agents
- **Supervised learning** — move prediction, outcome prediction, and policy modeling
- **Game theory research** — opening move frequency, position evaluation, and balance analysis