- Determine how many misclassified positions are illegal according to the rules of chess. The results are obtained by running
python3 chesscog.report.analyze_misclassified_positions --results data://recognition --dataset test
andpython3 chesscog.report.analyze_misclassified_positions --results data://transfer_learning/recognition --dataset test
):- In the test set (342 samples total), 21 positions were misclassified, of which 1 was illegal according to the rules of chess.
- In the test set of the transfer learning dataset (27 samples total), 3 positions were misclassified, of which all 3 were illegal according to the rules of chess.
- Display a warning in the inference scripts (
chesscog.recognition.recognition
andchesscog.transfer_learning.recognition
) if the predicted position is illegal according to the rules of chess.
Note: FEN positions are validated using the chess library.