|
|
--- |
|
|
license: mit |
|
|
version: 1.0.0 |
|
|
language: |
|
|
- en |
|
|
pretty_name: Tic-Tac-Toe Distinct Legal Move Dataset |
|
|
size_categories: |
|
|
- 1M<n<10M |
|
|
tags: |
|
|
- tic-tac-toe |
|
|
- board-games |
|
|
- game-dataset |
|
|
- ai |
|
|
- ml |
|
|
- reinforcement-learning |
|
|
- supervised-learning |
|
|
- q-learning |
|
|
- minimax |
|
|
- explainable-ai |
|
|
- symmetry |
|
|
- dataset-generation |
|
|
- python |
|
|
- educational |
|
|
task_categories: |
|
|
- reinforcement-learning |
|
|
- tabular-classification |
|
|
- tabular-regression |
|
|
- other |
|
|
--- |
|
|
|
|
|
# 🧠 Tic-Tac-Toe Distinct Legal Move Dataset |
|
|
|
|
|
## 📘 Overview |
|
|
This dataset contains **all possible legal Tic-Tac-Toe game sequences** — that is, every unique path through the complete game tree (about **255,168 distinct sequences**) from start to finish. |
|
|
|
|
|
Each row represents one move in one possible game, with metadata like: |
|
|
- Which player made the move |
|
|
- The board state before the move |
|
|
- The move position |
|
|
- The final game result (Win/Loss/Draw) |
|
|
- A `symmetry_id` that groups equivalent boards under rotations and reflections |
|
|
|
|
|
This dataset can be used for **machine learning experiments**, **game theory analysis**, or **reinforcement learning** training. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧩 Dataset Structure |
|
|
|
|
|
| Column | Description | |
|
|
|:--------|:-------------| |
|
|
| `game_id` | Unique identifier for each full game sequence | |
|
|
| `step` | Step number within the game (1–9) | |
|
|
| `player` | `X` or `O` — the player who made the move | |
|
|
| `board_state` | The board represented as a 9-element list (1 = X, -1 = O, 0 = Empty) | |
|
|
| `next_move` | The index (0–8) of the next move in the flattened 3×3 grid | |
|
|
| `result` | The final outcome of the game (`X Win`, `O Win`, or `Draw`) | |
|
|
| `board_state_str` | Human-readable board string (e.g., `X,_,_,O,_,_,_,_,_`) | |
|
|
| `symmetry_id` | A hash of the canonical form of the board after accounting for rotations/reflections | |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔍 Example |
|
|
|
|
|
| game_id | step | player | board_state | next_move | result | board_state_str | symmetry_id | |
|
|
|:--------|:------|:--------|:--------------|:------------|:--------|:------------------|:---------------| |
|
|
| 1 | 1 | X | `[0,0,0,0,0,0,0,0,0]` | 0 | X Win | `_,_,_,_,_,_,_,_,_` | 812347823 | |
|
|
| 1 | 2 | O | `[1,0,0,0,0,0,0,0,0]` | 4 | X Win | `X,_,_,_,_,_,_,_,_` | 982734987 | |
|
|
| 1 | 3 | X | `[1,0,0,0,-1,0,0,0,0]` | 8 | X Win | `X,_,_,_,O,_,_,_,_` | 123459812 | |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧮 Data Generation Logic |
|
|
|
|
|
The dataset was programmatically generated by exploring the **complete Tic-Tac-Toe game tree** using recursive backtracking in Python. |
|
|
|
|
|
Key features: |
|
|
- **Symmetry normalization:** Boards that are identical under rotation/reflection are reduced to a single canonical form. |
|
|
- **Comprehensive coverage:** Every legal sequence is included. |
|
|
- **Direct serialization:** Each move is written as a CSV row for efficient analysis. |
