File size: 8,806 Bytes
162232a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 |
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "3ca08817",
"metadata": {},
"outputs": [],
"source": [
"# !pip install seqeval"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c5958200",
"metadata": {},
"outputs": [],
"source": [
"# import torch\n",
"# torch.cuda.is_available(), torch.cuda.device_count()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "590c3f48",
"metadata": {},
"outputs": [],
"source": [
"import warnings\n",
"warnings.filterwarnings('ignore')\n",
"\n",
"import pickle\n",
"import numpy as np\n",
"import transformers\n",
"from transformers import Trainer\n",
"from datasets import load_metric\n",
"from datasets import load_dataset\n",
"from transformers import AutoTokenizer\n",
"from transformers import TrainingArguments\n",
"from transformers import AutoModelForTokenClassification\n",
"from transformers import DataCollatorForTokenClassification"
]
},
{
"cell_type": "markdown",
"id": "44d7c35c",
"metadata": {},
"source": [
"## Helpful funcs "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5c9e36d9",
"metadata": {},
"outputs": [],
"source": [
"def align_labels_with_tokens(labels: list, word_ids: list) -> list:\n",
" \"\"\"\n",
" Repeat label for each splitted token\n",
"\n",
" :param labels: list of entities token\n",
" :type labels: list\n",
" :param word_ids: list of word ids (repeadted if word was splitted)\n",
" :type word_ids: list\n",
" :return: list of aligned labels for tokenized sequence\n",
" :rtype: list\n",
" \"\"\"\n",
" return [-100 if i is None else labels[i] for i in word_ids]\n",
"\n",
"def tokenize_and_align_labels(examples):\n",
" \"\"\"\n",
" Tokenizing input sequence with corresponding labels\n",
"\n",
" :param examples: DatasetDict object with sequences and label ids\n",
" :type examples: DatasetDict\n",
" :return: DatasetDict with tokenizer output\n",
" :rtype: DatasetDict\n",
" \"\"\"\n",
" tokenized_inputs = tokenizer(\n",
" examples[\"sequences\"], truncation=True, is_split_into_words=True\n",
" )\n",
" all_labels = examples[\"ids\"]\n",
" new_labels = []\n",
" for i, labels in enumerate(all_labels):\n",
" word_ids = tokenized_inputs.word_ids(i)\n",
" new_labels.append(align_labels_with_tokens(labels, word_ids))\n",
"\n",
" tokenized_inputs[\"labels\"] = new_labels\n",
" return tokenized_inputs\n",
"\n",
"def compute_metrics(eval_preds):\n",
" \"\"\"\n",
" Function for evaluate model\n",
" \n",
" :param eval_preds: model output\n",
" :type eval_preds: \n",
" \"\"\"\n",
" logits, labels = eval_preds\n",
" predictions = np.argmax(logits, axis=-1)\n",
"\n",
" # Remove ignored index (special tokens) and convert to labels\n",
" true_labels = [[label_names[l] for l in label if l != -100] for label in labels]\n",
" true_predictions = [[label_names[p] for (p, l) in zip(prediction, label) if l != -100]\n",
" for prediction, label in zip(predictions, labels)\n",
" ]\n",
" all_metrics = metric.compute(predictions=true_predictions, references=true_labels)\n",
" return {\n",
" \"precision\": all_metrics[\"overall_precision\"],\n",
" \"recall\": all_metrics[\"overall_recall\"],\n",
" \"f1\": all_metrics[\"overall_f1\"],\n",
" \"accuracy\": all_metrics[\"overall_accuracy\"],\n",
" }"
]
},
{
"cell_type": "markdown",
"id": "8760e709",
"metadata": {},
"source": [
"## Load Data"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e8c723f7",
"metadata": {},
"outputs": [],
"source": [
"raw_datasets = load_dataset(\"surdan/nerel_short\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e540a898",
"metadata": {},
"outputs": [],
"source": [
"raw_datasets"
]
},
{
"cell_type": "markdown",
"id": "5a4947d1",
"metadata": {},
"source": [
"## Preprocess data"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8829557e",
"metadata": {},
"outputs": [],
"source": [
"model_checkpoint = \"cointegrated/LaBSE-en-ru\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b6c13ad1",
"metadata": {},
"outputs": [],
"source": [
"tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ea2c1a9e",
"metadata": {},
"outputs": [],
"source": [
"tokenized_datasets = raw_datasets.map(\n",
" tokenize_and_align_labels,\n",
" batched=True,\n",
" remove_columns=raw_datasets[\"train\"].column_names,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b15c3cf1",
"metadata": {},
"outputs": [],
"source": [
"tokenized_datasets"
]
},
{
"cell_type": "markdown",
"id": "e9b5b9b1",
"metadata": {},
"source": [
"## Init Training pipeline"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b24d86e3",
"metadata": {},
"outputs": [],
"source": [
"with open('id_to_label_map.pickle', 'rb') as f:\n",
" map_id_to_label = pickle.load(f)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1d90a6d9",
"metadata": {},
"outputs": [],
"source": [
"data_collator = DataCollatorForTokenClassification(tokenizer=tokenizer)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d890df2",
"metadata": {},
"outputs": [],
"source": [
"id2label = {str(k): v for k, v in map_id_to_label.items()}\n",
"label2id = {v: k for k, v in id2label.items()}\n",
"label_names = list(id2label.values())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "31bcfd6c",
"metadata": {},
"outputs": [],
"source": [
"model = AutoModelForTokenClassification.from_pretrained(\n",
" model_checkpoint,\n",
" id2label=id2label,\n",
" label2id=label2id,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "84497580",
"metadata": {},
"outputs": [],
"source": [
"model.config.num_labels"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1ccfbf74",
"metadata": {},
"outputs": [],
"source": [
"args = TrainingArguments(\n",
" \"LaBSE_ner_nerel\",\n",
" evaluation_strategy=\"epoch\",\n",
" save_strategy=\"no\",\n",
" learning_rate=2e-5,\n",
" num_train_epochs=25,\n",
" weight_decay=0.01,\n",
" push_to_hub=False,\n",
" per_device_train_batch_size = 4 ## depending on the total volume of memory of your GPU\n",
")"
]
},
{
"cell_type": "markdown",
"id": "c798d567",
"metadata": {},
"source": [
"## Train model"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1348d188",
"metadata": {},
"outputs": [],
"source": [
"## for compute_metrics function\n",
"metric = load_metric(\"seqeval\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5cff0367",
"metadata": {},
"outputs": [],
"source": [
"trainer = Trainer(\n",
" model=model,\n",
" args=args,\n",
" train_dataset=tokenized_datasets[\"train\"],\n",
" eval_dataset=tokenized_datasets[\"dev\"],\n",
" data_collator=data_collator,\n",
" compute_metrics=compute_metrics,\n",
" tokenizer=tokenizer,\n",
")\n",
"trainer.train()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "576a10f4",
"metadata": {},
"outputs": [],
"source": [
"trainer.save_model(\"LaBSE_nerel_last_checkpoint\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "451d6db1",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "hf_env",
"language": "python",
"name": "hf_env"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.10"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|