tokenizer.py

"""
Custom Chess Tokenizer for the Chess Challenge.

This tokenizer treats each move as a single token using the extended UCI notation
from the Lichess dataset (e.g., WPe2e4, BNg8f6).

The dataset format uses:
- W/B prefix for White/Black
- Piece letter: P=Pawn, N=Knight, B=Bishop, R=Rook, Q=Queen, K=King
- Source and destination squares (e.g., e2e4)
- Special suffixes: (x)=capture, (+)=check, (+*)=checkmate, (o)/(O)=castling
"""

from __future__ import annotations

import json
import os
from pathlib import Path
from typing import Dict, List, Optional


from transformers import PreTrainedTokenizer

class ChessTokenizer(PreTrainedTokenizer):
    model_input_names = ["input_ids", "attention_mask"]
    
    # Special tokens
    PAD_TOKEN = "[PAD]"
    BOS_TOKEN = "[BOS]" # Beginning of Sequence (Start of Game)
    EOS_TOKEN = "[EOS]" # End of Sequence (End of Game)
    UNK_TOKEN = "[UNK]"

    vocab_files_names = {
        "vocab_file": "vocab.json"
    }
    
    def __init__(
        self,
        vocab_file: Optional[str] = None,
        vocab: Optional[Dict[str, int]] = None,
        **kwargs,
    ):
        self._pad_token = self.PAD_TOKEN
        self._bos_token = self.BOS_TOKEN
        self._eos_token = self.EOS_TOKEN
        self._unk_token = self.UNK_TOKEN
        
        # Clean kwargs to avoid conflicts
        kwargs.pop("pad_token", None)
        kwargs.pop("bos_token", None)
        kwargs.pop("eos_token", None)
        kwargs.pop("unk_token", None)
        
        
        self.vocab_file = vocab_file
        
        if vocab is not None:
            self._vocab = vocab
        elif vocab_file is not None and os.path.exists(vocab_file):
            with open(vocab_file, "r", encoding="utf-8") as f:
                self._vocab = json.load(f)
        else:
            self._vocab = self._create_default_vocab()
            
        self._ids_to_tokens = {v: k for k, v in self._vocab.items()}
        
        super().__init__(
            pad_token=self._pad_token,
            bos_token=self._bos_token,
            eos_token=self._eos_token,
            unk_token=self._unk_token,
            **kwargs,
        )

    def _create_default_vocab(self) -> Dict[str, int]:
        """Creates basic vocab. Use build_vocab_from_dataset for full vocab."""
        # 4 Special + 12 Pieces + 64 Squares = 80 tokens total
        special = [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]
        vocab = {t: i for i, t in enumerate(special)}
        return vocab

    @property
    def vocab_size(self) -> int:
        return len(self._vocab)

    def get_vocab(self) -> Dict[str, int]:
        return dict(self._vocab)

    def _tokenize(self, text: str) -> List[str]:
        """
        Splits text "WPe2e4 BNg8f6" into ["WP", "e2", "e4", "BN", "g8", "f6"]
        """
        tokens = []
        # Split by space to get individual moves first
        raw_moves = text.strip().split()

        
        for move in raw_moves:
            # Check length to ensure it's a valid move string
            if len(move) >= 6:
                # Part 1: Player + Piece (Indices 0-2, e.g., "WP")
                tokens.append(move[:2])
                
                # Part 2: Start Square (Indices 2-4, e.g., "e2")
                tokens.append(move[2:4])
                
                # Part 3: End Square (Indices 4-6, e.g., "e4")
                tokens.append(move[4:])
                
                # Note: Suffixes like (x) or promotions (=Q) are ignored 
                # in this strict 3-token split implementation.

            else:
                tokens.append(self.UNK_TOKEN)


        return tokens

    def _convert_token_to_id(self, token: str) -> int:
        return self._vocab.get(token, self._vocab.get(self.UNK_TOKEN))

    def _convert_id_to_token(self, index: int) -> str:
        return self._ids_to_tokens.get(index, self.UNK_TOKEN)

    def convert_tokens_to_string(self, tokens: List[str]) -> str:
        """
        Reconstructs the move string. 
        Note: This simply joins them. You might need custom logic 
        if you want to strictly recreate 'WPe2e4' from ['WP','e2','e4'].
        """
        return " ".join(t for t in tokens if t not in [
            self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN
        ])

    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> tuple:
        if not os.path.isdir(save_directory):
            os.makedirs(save_directory, exist_ok=True)
        
        vocab_file = os.path.join(
            save_directory, 
            (filename_prefix + "-" if filename_prefix else "") + "vocab.json"
        )
        
        with open(vocab_file, "w", encoding="utf-8") as f:
            json.dump(self._vocab, f, ensure_ascii=False, indent=2)
            
        return (vocab_file,)

    @classmethod
    def build_vocab_from_dataset(cls, dataset_name="dlouapre/lichess_2025-01_1M", split="train", max_samples=10000):
        """Scans dataset to find all unique pieces and squares."""
        from datasets import load_dataset
        dataset = load_dataset(dataset_name, split=split, streaming=True)
        
        pieces = set()
        squares = set()
        endings = set()
        
        print("Building vocabulary...")
        count = 0
        for example in dataset:
            moves = example["text"].split()
            for move in moves:
                if len(move) >= 6:
                    pieces.add(move[:2])    # WP, BN, etc.
                    squares.add(move[2:4])  # e2
                    squares.add(move[4:])  # e4


            count += 1
            if count >= max_samples:
                break

        # Combine into vocab structure
        special = [cls.PAD_TOKEN, cls.BOS_TOKEN, cls.EOS_TOKEN, cls.UNK_TOKEN]
        all_tokens = special + sorted(list(pieces)) + sorted(list(endings)) + sorted(list(squares))
        
        vocab = {token: idx for idx, token in enumerate(all_tokens)}
        return cls(vocab=vocab)