Create validate_data.py

validate_data.py

@@ -0,0 +1,58 @@
+"""Validation functions for the Chessbench dataset.
+
+This script validates the integrity of the Chessbench dataset by checking that each record adheres to expected formats and value ranges.
+It ensures that FEN strings and move UCI strings are valid, and that win probabilities and mate labels are consistent.
+"""
+
+import chess
+from tqdm import tqdm
+
+from athena.datasets.chessbenchmate.dataset import ChessbenchDataset
+
+
+def validate_record(record):
+    fen, move, win_prob, mate = record
+    assert isinstance(fen, str)
+    assert isinstance(move, str)
+
+    # Check for valid fen
+    board = chess.Board(fen)
+    assert board.is_valid(), f"Invalid FEN: {fen}"
+
+    # Check for valid move
+    uci_move = chess.Move.from_uci(move)
+    assert uci_move in board.legal_moves, f"Illegal move {move} for board {fen}"
+
+    if mate == "#":
+        assert win_prob == 1.0, f"Unexpected win_prob {win_prob} for mate {mate}"
+    elif isinstance(mate, int):
+        # If mate is "#" or an integer, win_prob should be 1.0 or 0.0
+        assert mate != 0, "Mate cannot be zero"
+        assert win_prob in (0.0, 1.0), f"Unexpected win_prob {win_prob} for mate {mate}"
+        if mate > 0:
+            assert win_prob == 1.0, f"Unexpected win_prob {win_prob} for mate {mate}"
+        else:
+            assert win_prob == 0.0, f"Unexpected win_prob {win_prob} for mate {mate}"
+    elif mate == "-":
+        assert isinstance(win_prob, float) and 0.0 < win_prob < 1.0
+
+
+test_dataset = ChessbenchDataset(dir="src/athena/datasets/chessbenchmate/data", mode="test")
+train_dataset = ChessbenchDataset(dir="src/athena/datasets/chessbenchmate/data", mode="train")
+
+
+for i in tqdm(range(len(test_dataset))):
+    record = test_dataset[i]
+    try:
+        validate_record(record)
+    except AssertionError as e:
+        print(f"Validation error in test dataset at index {i}: {e}")
+        raise
+
+for i in tqdm(range(len(train_dataset))):
+    record = train_dataset[i]
+    try:
+        validate_record(record)
+    except AssertionError as e:
+        print(f"Validation error in train dataset at index {i}: {e}")
+        raise