tokenizer.py

# Version  (Player (Color + Piece), Source_S, Destination_D, Suffix)

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):
    """
    Sub-move tokenizer for chess moves using extended UCI notation.
    
    This tokenizer splits each move into atomic components:
        - Players (color + piece): WP, WN, WB, WR, WQ, WK, etc.
        - Source square: e2
        - Destination square: e4
        - Optional suffixes: x (capture), + (check), * (checkmate), o/O (castling)
    
    Example:
        Move "WPe2e4(x+)" -> ["WP", "e2_S", "e4_D", "(x+)"]
    """
    
    model_input_names = ["input_ids", "attention_mask"]
    vocab_files_names = {"vocab_file": "vocab.json"}
    
    # Special tokens
    PAD_TOKEN = "[PAD]"
    BOS_TOKEN = "[BOS]"
    EOS_TOKEN = "[EOS]"
    UNK_TOKEN = "[UNK]"
    
    # Atomic suffix tokens for default vocab
    SUFFIX_TOKENS = ["(x)", "(+)", "(*)", "(o)", "(O)", "(+*)", "(x+)"]
    
    def __init__(
        self,
        vocab_file: Optional[str] = None,
        vocab: Optional[Dict[str, int]] = None,
        **kwargs,
    ):
        # Special tokens
        self._pad_token = self.PAD_TOKEN
        self._bos_token = self.BOS_TOKEN
        self._eos_token = self.EOS_TOKEN
        self._unk_token = self.UNK_TOKEN

        # Remove duplicates from kwargs
        kwargs.pop("pad_token", None)
        kwargs.pop("bos_token", None)
        kwargs.pop("eos_token", None)
        kwargs.pop("unk_token", None)
        
        # Load or create vocab
        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()
        
        # Reverse mapping
        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]:
        """
        Build a fixed vocab based on chess grammar for sub-moves.
        Useful for predefined grammar instead of dataset-based vocab.
        """
        colors = ["W", "B"]
        pieces = ["P", "N", "B", "R", "Q", "K"]
        files = ["a", "b", "c", "d", "e", "f", "g", "h"]
        ranks = ["1", "2", "3", "4", "5", "6", "7", "8"]
        squares = [f + r for f in files for r in ranks]

        players = [c + p for c in colors for p in pieces]

        # Source and destination tokens
        sources = [sq + "_S" for sq in squares]
        dests = [sq + "_D" for sq in squares]

        # Build all possible sub-tokens
        vocab_tokens = players + sources + dests + self.SUFFIX_TOKENS

        # Add special tokens at the start
        special_tokens = [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]
        vocab = {token: idx for idx, token in enumerate(special_tokens + vocab_tokens)}
        return vocab
    
    def _tokenize(self, text: str) -> List[str]:
        """
        Convert a string of moves into sub-move tokens.
        """
        tokens: List[str] = []
        moves = text.strip().split()
        for move in moves:
            if not move:
                continue
            
            # Color + Piece
            tokens.append(move[:2])  # WP, BN, etc.

            # Source square with _S
            tokens.append(move[2:4] + "_S")

            # Destination square with _D
            tokens.append(move[4:6] + "_D")

            if (len(move)>6):
                tokens.append(move[6:])
            
        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:
        """Convert a list of tokens back to a string."""
        # Filter out special tokens for cleaner output
        special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
        clean_tokens = []
        for t in tokens:
            if t in special:
                continue
            # Remove everything from _ onward
            if "_" in t:
                clean_tokens.append(t.split("_")[0])
            else:
                clean_tokens.append(t)
                
        result = ""
        temp = "".join(token for token in clean_tokens)

        for i, str in enumerate(temp):
            if str in ["W", "B"]:
                if result == "":
                    result += str
                elif temp[i-1].isnumeric() or temp[i-1]==")":
                    result += " " + str
                else :  
                    result += str
            else :  
                result += str   
                           
        return result.split()[0]
    
    @property
    def vocab_size(self) -> int:
        return len(self._vocab)
    
    def get_vocab(self) -> Dict[str, int]:
        return dict(self._vocab)
    
    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: int = 1) -> "ChessTokenizer":
        """
        Build vocab from dataset iterator using sub-move tokens.
        """
        from collections import Counter
        token_counts = Counter()
        for game in iterator:
            sub_tokens = cls()._tokenize(game)
            token_counts.update(sub_tokens)
        tokens = [token for token, count in token_counts.items() if count >= min_frequency]
        tokens = sorted(tokens)
        special_tokens = [cls.PAD_TOKEN, cls.BOS_TOKEN, cls.EOS_TOKEN, cls.UNK_TOKEN]
        vocab = {token: idx for idx, token in enumerate(special_tokens + tokens)}
        return cls(vocab=vocab)
    
    @classmethod
    def build_vocab_from_dataset(
        cls,
        dataset_name: str = "dlouapre/lichess_2025-01_1M",
        split: str = "train",
        column: str = "text",
        min_frequency: int = 500,
        max_samples: Optional[int] = 100000,
    ) -> "ChessTokenizer":
        from datasets import load_dataset
        dataset = load_dataset(dataset_name, split=split)
        if max_samples is not None:
            dataset = dataset.select(range(min(max_samples, len(dataset))))
        def game_iterator():
            for example in dataset:
                yield example[column]
        return cls.build_vocab_from_iterator(game_iterator(), min_frequency=min_frequency)


def count_vocab_from_dataset(
    dataset_name: str = "dlouapre/lichess_2025-01_1M",
    split: str = "train",
    column: str = "text",
    max_samples: Optional[int] = 10000,
) -> Dict[str, int]:
    """
    Count sub-move token frequencies in a dataset (useful for vocab analysis).
    """
    from collections import Counter
    from datasets import load_dataset
    
    dataset = load_dataset(dataset_name, split=split)
    if max_samples is not None:
        dataset = dataset.select(range(min(max_samples, len(dataset))))
    
    token_counts = Counter()
    for example in dataset:
        moves = example[column].strip().split()
        # Use sub-tokenization
        tokenizer = ChessTokenizer()
        for move in moves:
            token_counts.update(tokenizer._tokenize(move))
    return dict(token_counts)