The dataset for this competition was constructed using [Othello Games](https://www.kaggle.com/datasets/andrefpoliveira/othello-games) - a dataset by @andrefpoliveira. Board states were reconstructed from move sequences in the dataset, and evaluated using [Edax](https://github.com/abulmo/edax-reversi) @ depth=12. 

Board states are provided as flat, length-64 vectors of integers, with the board size being 8x8. An example has been visualized below.

![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F8509053%2Fd6fe93ae5d1a88e5c5e154ed8f1fdb4b%2FExampleBoard.png?generation=1753290071132419&alt=media)

## Files

*   **train.csv** - the training set, containing 6,400 pairs of board states and evaluation scores
*   **test.csv** - the test set, containing 1,600 board states for which to predict evaluation scores
*   **sample_submission.csv** - a sample submission file in the correct format

## Columns

*   `id` - identifier for the record in the dataset
*   `turn_player_advantage` - an evaluation score, measured (in discs) relative to the turn player
*   `cell_[0-63]` - the 64 cells of the grid, row-wise, in which pieces are denoted as: 1=black, 0=empty, -1=white

## External Dataset

You may wish to supplement the training data with the [Othello Board Evaluations](https://www.kaggle.com/datasets/petermoorhouse/othello-board-evaluations) dataset. The competition data was constructed in the same manner as this dataset, but with board states sourced from an entirely separate set of games.