tokenizer.py

"""
Custom Chess Tokenizer - Decomposed Move Representation.

Decomposes chess moves into components: Color, Piece, From square, To square, Modifiers.
Uses a fixed vocabulary of ~88 tokens.
"""

from __future__ import annotations

import json
import os
import re
from typing import Dict, List, Optional

from transformers import PreTrainedTokenizer


class ChessTokenizer(PreTrainedTokenizer):
    model_input_names = ["input_ids", "attention_mask"]
    
    PAD_TOKEN = "[PAD]"
    BOS_TOKEN = "[BOS]"
    EOS_TOKEN = "[EOS]"
    UNK_TOKEN = "[UNK]"
    
    COLORS = ["[W]", "[B]"]
    PIECES = ["P", "N", "B", "R", "Q", "K"]
    SQUARES = [f + r for f in "abcdefgh" for r in "12345678"]
    MODIFIERS = ["x", "+", "#", "+*", "=Q", "=R", "=B", "=N", "O-O", "O-O-O", "o", "O"]
    
    MOVE_PATTERN = re.compile(
        r'^([WB])([PNBRQK])([a-h][1-8])([a-h][1-8])(=[QRBN])?(\([xoO+*]+\))?$'
    )
    
    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

        kwargs.pop("pad_token", None)
        kwargs.pop("bos_token", None)
        kwargs.pop("eos_token", None)
        kwargs.pop("unk_token", None)

        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]:
        tokens = [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]
        tokens.extend(self.COLORS)
        tokens.extend(self.PIECES)
        tokens.extend(self.SQUARES)
        tokens.extend(self.MODIFIERS)
        return {token: idx for idx, token in enumerate(tokens)}
    
    def _parse_move(self, move: str) -> List[str]:
        """Parse a single move into component tokens."""
        match = self.MOVE_PATTERN.match(move)
        if not match:
            return [self.UNK_TOKEN]
        
        tokens = [f"[{match.group(1)}]", match.group(2), match.group(3), match.group(4)]
        
        if match.group(5):
            tokens.append(match.group(5))
        
        if match.group(6):
            suffix_content = match.group(6)[1:-1]
            if "x" in suffix_content:
                tokens.append("x")
            if "+*" in suffix_content:
                tokens.append("+*")
            elif "+" in suffix_content:
                tokens.append("+")
            if suffix_content == "o":
                tokens.append("o")
            elif suffix_content == "O":
                tokens.append("O")
        
        return tokens
    
    def _tokenize(self, text: str) -> List[str]:
        """Tokenize a string of moves into component tokens."""
        tokens = []
        for move in text.strip().split():
            tokens.extend(self._parse_move(move))
        return tokens
    
    def _convert_token_to_id(self, token: str) -> int:
        return self._vocab.get(token, self._vocab.get(self.UNK_TOKEN, 0))
    
    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:
        """Reconstruct moves from component tokens."""
        special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
        result = []
        current_move = []
        
        for token in tokens:
            if token in special:
                if current_move:
                    result.append(self._reconstruct_move(current_move))
                    current_move = []
                continue
            
            current_move.append(token)
            if self._is_complete_move(current_move):
                result.append(self._reconstruct_move(current_move))
                current_move = []
        
        if current_move:
            result.append(self._reconstruct_move(current_move))
        
        return " ".join(result)
    
    def _is_complete_move(self, tokens: List[str]) -> bool:
        """Check if tokens form a complete move."""
        if len(tokens) < 4:
            return False
        
        if (tokens[0] in self.COLORS and tokens[1] in self.PIECES and
            tokens[2] in self.SQUARES and tokens[3] in self.SQUARES):
            if len(tokens) == 4:
                return True
            for t in tokens[4:]:
                if t in self.COLORS or (t not in self.MODIFIERS and not t.startswith("=")):
                    return True
            return True
        return False
    
    def _reconstruct_move(self, tokens: List[str]) -> str:
        """Reconstruct a move string from component tokens."""
        if not tokens or len(tokens) < 4:
            return "".join(tokens)
        
        color = tokens[0][1] if tokens[0] in self.COLORS else tokens[0]
        move = color + "".join(tokens[1:4])
        
        suffixes = []
        for t in tokens[4:]:
            if t.startswith("="):
                move += t
            elif t in ["x", "+", "+*", "o", "O"]:
                suffixes.append(t)
        
        if suffixes:
            move += "(" + "".join(suffixes) + ")"
        
        return move
    
    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_iterator(cls, iterator, min_frequency=1):
        return cls()
    
    @classmethod
    def build_vocab_from_dataset(cls, **kwargs):
        return cls()

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