File size: 13,155 Bytes
ec24fd5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 |
"""
This script defines a custom tokenizer, `SupplyChainTokenizer`, specifically designed
for a collaborative predictive supply chain model using Transformer-based
architecture. It leverages a custom, industry-specific vocabulary (loaded from
`vocab.json`) to prioritize domain-relevant tokens (SKUs, store IDs, plant IDs,
promotion types, etc.) while employing Byte-Pair Encoding (BPE) to handle
out-of-vocabulary words and variations.
The script also includes a comprehensive example usage section demonstrating
how to create, train, use, save, and load the tokenizer. This tokenizer is a
critical component for bridging the gap between raw supply chain data and
a Transformer-based forecasting model.
"""
import json
import os
from typing import List, Dict, Union, Tuple
from tokenizers import (
Tokenizer,
models,
normalizers,
pre_tokenizers,
decoders,
trainers,
processors,
)
from tokenizers.pre_tokenizers import WhitespaceSplit, Digits
from tokenizers import Regex
import pandas as pd
class SupplyChainTokenizer:
"""
A custom tokenizer designed for the Enhanced Business Model for Collaborative
Predictive Supply Chain. It prioritizes industry-specific tokens from a
`vocab.json` file and uses Byte-Pair Encoding (BPE) for out-of-vocabulary
(OOV) words. It handles various data types expected in supply chain data.
Args:
vocab_path (str): Path to the `vocab.json` file.
max_length (int, optional): Maximum sequence length. Defaults to 512.
"""
def __init__(self, vocab_path: str, max_length: int = 512):
if not os.path.exists(vocab_path):
raise FileNotFoundError(f"Vocabulary file not found: {vocab_path}")
self.vocab_path = vocab_path
self.max_length = max_length
# Load the custom vocabulary
with open(self.vocab_path, "r", encoding="utf-8") as f:
self.vocab = json.load(f)
# 1. Create the BPE model
self.bpe_model = models.BPE(
vocab=self.vocab, # Initialize with the custom vocabulary
merges=[], # We'll populate merges during training
unk_token="[UNK]", # Unknown token
)
# 2. Create a Tokenizer instance
self.tokenizer = Tokenizer(self.bpe_model)
# 3. Normalization (Lowercase and Unicode normalization)
self.tokenizer.normalizer = normalizers.Sequence(
[normalizers.NFD(), normalizers.Lowercase(), normalizers.StripAccents()]
)
# 4. Pre-tokenization (Splitting into words)
self.tokenizer.pre_tokenizer = pre_tokenizers.Sequence(
[WhitespaceSplit(), Digits(individual_digits=True)]
)
# 5. Decoder (Convert token IDs back to strings)
self.tokenizer.decoder = decoders.BPEDecoder()
# 6. Post-processing (Special tokens)
self.tokenizer.post_processor = processors.TemplateProcessing(
single="[CLS] $A [SEP]",
pair="[CLS] $A [SEP] $B:1 [SEP]:1",
special_tokens=[("[CLS]", self.vocab["[CLS]"]), ("[SEP]", self.vocab["[SEP]"])],
)
# Adding this, although not used in encode or encode_as_ids
self.pad_token_id = self.vocab["[PAD]"]
def train_bpe(self, files: Union[str, List[str]], vocab_size: int = 30000):
"""
Trains the BPE model on text files. This updates the `merges` of the
BPE model. This is *crucial* for handling words not in the initial
`vocab.json`.
Args:
files (Union[str, List[str]]): Path(s) to text file(s) for training.
vocab_size (int): The desired vocabulary size (including special tokens
and initial vocabulary).
"""
if isinstance(files, str):
files = [files]
# Create a trainer
trainer = trainers.BpeTrainer(
vocab_size=vocab_size,
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
initial_alphabet=pre_tokenizers.ByteLevel.alphabet(), # All single bytes
show_progress=True,
)
# Train the tokenizer
self.tokenizer.train(files, trainer=trainer)
def encode(self, text: str, text_pair: str = None) -> List[str]:
"""
Encodes text into a list of tokens.
Args:
text (str): The input text.
text_pair (str, optional): An optional second input string.
Returns:
List[str]: A list of tokens.
"""
encoded = self.tokenizer.encode(text, text_pair)
return encoded.tokens
def encode_as_ids(self, text: str, text_pair: str = None) -> List[int]:
"""
Encodes text into a list of token IDs.
Args:
text (str): The input text.
text_pair (str, optional): An optional second input string.
Returns:
List[int]: A list of token IDs.
"""
encoded = self.tokenizer.encode(text, text_pair)
return encoded.ids
def decode(self, ids: List[int], skip_special_tokens: bool = True) -> str:
"""
Decodes a list of token IDs back into a string.
Args:
ids (List[int]): The list of token IDs.
skip_special_tokens (bool): Whether to skip special tokens in decoding.
Returns:
str: The decoded string.
"""
return self.tokenizer.decode(ids, skip_special_tokens=skip_special_tokens)
def token_to_id(self, token: str) -> int:
"""
Converts a token to its corresponding ID.
Args:
token (str): The token.
Returns:
int: The token ID. Returns None if the token is not in the vocabulary.
"""
return self.vocab.get(token, self.vocab.get("[UNK]"))
def id_to_token(self, id_: int) -> str:
"""
Converts a token ID to its corresponding token.
Args:
id_ (int): The token ID.
Returns:
str: The token. Returns "[UNK]" if the ID is not in the vocabulary.
"""
# Reverse lookup (efficient if needed frequently)
reverse_vocab = {v: k for k, v in self.vocab.items()}
return reverse_vocab.get(id_, "[UNK]")
def get_vocab_size(self) -> int:
"""Gets the vocabulary size."""
return len(self.vocab)
def save(self, directory: str, prefix: str = None):
"""
Saves the tokenizer configuration and vocabulary to a directory.
Args:
directory (str): The directory to save to.
prefix (str, optional): An optional prefix for the filenames.
"""
if not os.path.exists(directory):
os.makedirs(directory)
# Save the tokenizer configuration
self.tokenizer.save(os.path.join(directory, (prefix + "-" if prefix else "") + "tokenizer.json"))
# Save a copy of the vocabulary (for easy access)
with open(os.path.join(directory, (prefix + "-" if prefix else "") + "vocab.json"), "w", encoding="utf-8") as f:
json.dump(self.vocab, f, ensure_ascii=False, indent=4)
@staticmethod
def from_pretrained(directory: str, prefix: str = None):
"""
Loads a pre-trained tokenizer from a directory.
Args:
directory (str): The directory to load from.
prefix (str, optional): The optional prefix used when saving.
Returns:
SupplyChainTokenizer: The loaded tokenizer.
"""
vocab_path = os.path.join(directory, (prefix + "-" if prefix else "") + "vocab.json")
# You could load the tokenizer.json, but since we have a custom class
# with training logic, it's better to reconstruct the object this way.
tokenizer = SupplyChainTokenizer(vocab_path)
tokenizer.tokenizer = Tokenizer.from_file(os.path.join(directory, (prefix + "-" if prefix else "") + "tokenizer.json"))
return tokenizer
def prepare_for_model(self, data: pd.DataFrame) -> Tuple[List[List[int]], List[List[int]]]:
"""
Prepares a Pandas DataFrame for the Transformer model. This is the
key method that integrates the tokenizer with the data.
Args:
data (pd.DataFrame): The input DataFrame, expected to have columns
like 'timestamp', 'sku', 'store_id', 'quantity', 'price',
'discount', 'promotion_id', etc. The exact columns depend on
the features you're using.
Returns:
Tuple[List[List[int]], List[List[int]]]: A tuple.
1. input_ids: List of token ID sequences for the model.
2. attention_mask: List of attention masks (1 for real tokens, 0 for padding).
"""
input_ids = []
attention_masks = []
for _, row in data.iterrows():
# Build the input string. This is where you define *how* your
# features are combined into a single sequence.
input_string = (
f"[CLS] timestamp: {row['timestamp']} "
f"sku: {row['sku']} store_id: {row['store_id']} "
f"quantity: {row['quantity']} price: {row['price']} "
f"discount: {row['discount']} "
)
# Add promotion information if available
if 'promotion_id' in row and not pd.isna(row['promotion_id']):
input_string += f"promotion_id: {row['promotion_id']} "
# Add any other relevant features here
if 'product_category' in row:
input_string += f"product_category: {row['product_category']} "
# Add other external features
input_string += "[SEP]"
# Tokenize
encoded = self.tokenizer.encode(input_string)
token_ids = encoded.ids
attention_mask = encoded.attention_mask
# Padding (up to max_length)
padding_length = self.max_length - len(token_ids)
if padding_length > 0:
token_ids += [self.pad_token_id] * padding_length
attention_mask += [0] * padding_length
elif padding_length < 0: # Truncation
token_ids = token_ids[:self.max_length]
attention_mask = attention_mask[:self.max_length]
input_ids.append(token_ids)
attention_masks.append(attention_mask)
return input_ids, attention_masks
# Example Usage (Illustrative)
if __name__ == "__main__":
# --- Create a dummy vocab.json ---
vocab = {
"[UNK]": 0,
"[CLS]": 1,
"[SEP]": 2,
"[PAD]": 3,
"[MASK]": 4,
"timestamp:": 5,
"sku:": 6,
"store_id:": 7,
"quantity:": 8,
"price:": 9,
"discount:": 10,
"promotion_id:": 11,
"product_category:": 12,
"SKU123": 13, # Example SKU
"SKU123-RED": 14, # Example SKU variant
"SKU123-BLUE": 15,
"STORE456": 16, # Example store ID
"PLANT789": 17, # Example plant ID
"WHOLESALER001": 18, # Example Wholesaler
"RETAILER002": 19, # Example Retailer
"BOGO": 20,
"DISCOUNT":21,
}
with open("vocab.json", "w") as f:
json.dump(vocab, f, indent=4)
# --- Create the tokenizer ---
tokenizer = SupplyChainTokenizer(vocab_path="vocab.json")
# --- Example training (on a dummy text file) ---
with open("training_data.txt", "w", encoding="utf-8") as f:
f.write("This is some example text for training the BPE model.\n")
f.write("SKU123 is a product. STORE456 is another. plant789 is, too.\n")
f.write("This file contains words not in the initial vocabulary.\n")
tokenizer.train_bpe("training_data.txt", vocab_size=50) # Small vocab for the example
# --- Example encoding ---
text = "timestamp: 2024-07-03 sku: SKU123 store_id: STORE456 quantity: 2 price: 10.99 discount: 0.0"
encoded_tokens = tokenizer.encode(text)
encoded_ids = tokenizer.encode_as_ids(text)
print(f"Encoded tokens: {encoded_tokens}")
print(f"Encoded IDs: {encoded_ids}")
decoded_text = tokenizer.decode(encoded_ids)
print(f"Decoded text: {decoded_text}")
# -- Example with DataFrame ---
data = {
'timestamp': ['2024-07-03 10:00:00', '2024-07-03 11:00:00'],
'sku': ['SKU123', 'SKU123-RED'],
'store_id': ['STORE456', 'STORE456'],
'quantity': [2, 1],
'price': [10.99, 12.99],
'discount': [0.0, 1.0],
'promotion_id': ['BOGO', None],
'product_category': ['Electronics', 'Electronics']
}
df = pd.DataFrame(data)
input_ids, attention_masks = tokenizer.prepare_for_model(df)
print(f"Input IDs (for model): {input_ids}")
print(f"Attention Masks: {attention_masks}")
# --- Save and load ---
tokenizer.save("my_tokenizer")
loaded_tokenizer = SupplyChainTokenizer.from_pretrained("my_tokenizer")
print(f"Loaded tokenizer vocab size: {loaded_tokenizer.get_vocab_size()}")
# Clean up example files
os.remove("vocab.json")
os.remove("training_data.txt") |