File size: 6,669 Bytes
bb88ae7 | 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 | """
Atomic Chess Tokenizer.
Decomposes chess moves into atomic components:
[Piece] + [Source] + [Destination] + [Suffix]
Example: "WPe2e4(x)" -> ["WP", "e2", "e4", "(x)"]
Benefits:
- Drastically reduces vocab size (~1200 -> ~90)
- Saves ~140k parameters in the embedding layer
- Allows the model to learn spatial relationships (e2 is close to e3)
"""
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"]
# Special tokens
PAD_TOKEN = "[PAD]"
BOS_TOKEN = "[BOS]"
EOS_TOKEN = "[EOS]"
UNK_TOKEN = "[UNK]"
# Regex to parse the extended UCI format
# Groups: 1=Piece, 2=Source, 3=Dest, 4=Suffix
MOVE_REGEX = re.compile(r"([WB][PNBRQK])([a-h][1-8])([a-h][1-8])(.*)")
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
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_atomic_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_atomic_vocab(self) -> Dict[str, int]:
"""
Manually builds the vocabulary because we know the rules of Chess.
We don't need to learn this from the dataset.
"""
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 (Color + Type)
colors = ['W', 'B']
pieces = ['P', 'N', 'B', 'R', 'Q', 'K']
for c in colors:
for p in pieces:
vocab[f"{c}{p}"] = idx
idx += 1
# 3. Squares (a1 to h8)
files = ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h']
ranks = ['1', '2', '3', '4', '5', '6', '7', '8']
for f in files:
for r in ranks:
vocab[f"{f}{r}"] = idx
idx += 1
# 4. Common Suffixes (derived from Lichess notation)
# (x)=capture, (+)=check, (#)=mate, (o)=castling
suffixes = ["(x)", "(+)", "(+*)", "(o)", "(O)", "=", "=Q", "=R", "=B", "=N"]
for s in suffixes:
vocab[s] = idx
idx += 1
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 a string of moves into atomic tokens.
"WPe2e4" -> ["WP", "e2", "e4"]
"""
raw_moves = text.strip().split()
tokens = []
for move in raw_moves:
match = self.MOVE_REGEX.match(move)
if match:
# Add piece, source, dest
tokens.extend([match.group(1), match.group(2), match.group(3)])
# Add suffix if it exists
suffix = match.group(4)
if suffix:
tokens.append(suffix)
else:
# Fallback for weird formatting (or UNK)
tokens.append(move)
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 moves from atomic tokens.
This is tricky because we need to join them without spaces,
but add spaces between actual moves.
"""
out = []
current_move = []
special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
for t in tokens:
if t in special:
continue
current_move.append(t)
# Logic to decide when a move ends
# A move usually ends after a Suffix OR after a Destination square if no suffix follows
# This heuristic is simple: if we have a piece, src, and dest, check next token
# Simplified reconstruction:
# Just join everything and use a heuristic to insert spaces?
# Better: The model generates atomic tokens.
# We know a move starts with [WB][PNBRQK].
# Robust reconstruction approach:
full_str = "".join([t for t in tokens if t not in special])
# Insert space before every Piece token (except the first one)
# Regex lookbehind isn't strictly necessary, we can just replace
formatted = re.sub(r'(?<!^)([WB][PNBRQK])', r' \1', full_str)
return formatted
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,)
# We don't really need build_vocab_from_dataset anymore as we hardcoded the rules,
# but we keep the method signature to satisfy the template.
@classmethod
def build_vocab_from_dataset(cls, *args, **kwargs):
print("Note: Atomic tokenizer uses a static vocabulary rule set.")
return cls()
|