File size: 6,551 Bytes
c8046f8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
"""
Custom Chess Tokenizer for the Chess Challenge.

This tokenizer decomposes moves into atomic tokens:
Piece -> Source Square -> Target Square -> Suffixes.
Example: "WPe2e4" -> ['P', 'e2', 'e4'] (Color is implicit to save context)
Example: "Bxb7+" -> ['B', 'c8', 'b7', '(x)', '(+)']
"""

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"]
    vocab_files_names = {"vocab_file": "vocab.json"}
    
    # Special tokens
    PAD_TOKEN = "[PAD]"
    BOS_TOKEN = "[BOS]"
    EOS_TOKEN = "[EOS]"
    UNK_TOKEN = "[UNK]"
    
    # Atomic components
    PIECES = ["P", "N", "B", "R", "Q", "K"]
    FILES = "abcdefgh"
    RANKS = "12345678"
    SUFFIXES = ["(x)", "(+)", "(+*)", "(o)", "(O)", "(=)"]

    def __init__(
        self,
        vocab_file: Optional[str] = None,
        vocab: Optional[Dict[str, int]] = None,
        **kwargs,
    ):
        # Initialize 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

        # Clean kwargs
        kwargs.pop("pad_token", None)
        kwargs.pop("bos_token", None)
        kwargs.pop("eos_token", None)
        kwargs.pop("unk_token", None)
        
        # Load or create FIXED vocabulary
        if vocab is not None:
            self._vocab = vocab
        else:
            self._vocab = self._create_fixed_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_fixed_vocab(self) -> Dict[str, int]:
        """Creates the fixed vocabulary of ~80 atomic tokens."""
        vocab = {}
        idx = 0
        
        # 1. Special Tokens
        for token in [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]:
            vocab[token] = idx
            idx += 1
            
        # 2. Pieces
        for p in self.PIECES:
            vocab[p] = idx
            idx += 1
            
        # 3. Squares (a1...h8)
        # We treat squares as atomic tokens for better spatial learning
        for f in self.FILES:
            for r in self.RANKS:
                vocab[f"{f}{r}"] = idx
                idx += 1
                
        # 4. Suffixes
        for s in self.SUFFIXES:
            vocab[s] = idx
            idx += 1
            
        return vocab

    @classmethod
    def build_vocab_from_dataset(
        cls,
        dataset_name: str = "dlouapre/lichess_2025-01_1M",
        **kwargs
    ) -> "ChessTokenizer":
        """
        Override: Returns the tokenizer with the fixed vocabulary immediately.
        We do not need to scan the dataset anymore.
        """
        print("Initializing Fixed Vocabulary Tokenizer (Deconstructed Strategy)...")
        return cls()

    @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]:
        """
        Decomposes move strings into atomic tokens.
        Input: "[BOS] WPe2e4 BNg8f6"
        Output: ['[BOS]', 'P', 'e2', 'e4', 'N', 'g8', 'f6']
        """
        tokens = []
        moves = text.strip().split()
        
        # Set of special tokens for quick lookup
        special_tokens = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}

        for move in moves:
            # Skip empty strings
            if not move: continue
            
            # 1. Handle Special Tokens (Important for data.py compatibility)
            if move in special_tokens:
                tokens.append(move)
                continue

            # 2. Regex to parse Lichess format: WPe2e4(x)
            # Group 1: Color (W/B) - Ignored
            # Group 2: Piece (P/N/B/R/Q/K)
            # Group 3: Source (e.g. e2)
            # Group 4: Target (e.g. e4)
            # Group 5: Suffix (optional)
            match = re.match(r"([WB])([PNBRQK])([a-h][1-8])([a-h][1-8])(.*)", move)
            
            if match:
                _, piece, src, dst, suffix = match.groups()
                tokens.extend([piece, src, dst])
                if suffix:
                    if suffix in self._vocab:
                        tokens.append(suffix)
            else:
                # Fallback for unexpected formats
                found_any = False
                
                # Check for piece
                for p in self.PIECES:
                    if p in move:
                        tokens.append(p)
                        found_any = True
                        break
                
                # Check for squares
                squares = re.findall(r"[a-h][1-8]", move)
                tokens.extend(squares)
                if squares: found_any = True

                # Check for suffixes
                for s in self.SUFFIXES:
                    if s in move:
                        tokens.append(s)
                        found_any = True
                
                if not found_any:
                    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 a readable string (space separated for clarity)
        special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
        return " ".join(t for t in tokens if t not in special)

    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,)