Spaces:
Sleeping
Sleeping
File size: 9,850 Bytes
5bdc726 |
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 |
{
"cells": [
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"id": "S52EVP7k-rl7"
},
"outputs": [],
"source": [
"import pandas as pd\n",
"import torch\n",
"import re\n",
"import string\n",
"import numpy as np\n",
"import streamlit as st\n",
"import faiss # хранение индексов\n",
"from tqdm import tqdm\n",
"from transformers import AutoTokenizer, AutoModel\n",
"from joblib import dump, load # Для сохранения/загрузки эмбэддингов"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"id": "12BEEwcF-rl9"
},
"outputs": [],
"source": [
"path = '/content/movies_filtered.csv' # ИЗМЕНИ ТУТ ПУТЬ!\n",
"a\n",
"df = pd.read_csv(path)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"id": "df5lg8-m-rl-"
},
"outputs": [],
"source": [
"def clean(text):\n",
" text = text.lower() # Нижний регистр\n",
" text = re.sub(r'\\d+', ' ', text) # Удаляем числа\n",
" # text = text.translate(str.maketrans('', '', string.punctuation)) # Удаляем пунктуацию\n",
" text = re.sub(r'\\s+', ' ', text) # Удаляем лишние пробелы\n",
" text = text.strip() # Удаляем начальные и конечные пробелы\n",
" text = re.sub(r'\\s+|\\n', ' ', text) # Удаляет \\n и \\xa0\n",
" # text = re.sub(r'\\b\\w{1,2}\\b', '', text) # Удаляем слова длиной менее 3 символов\n",
" # Дополнительные шаги, которые могут быть полезны в данном контексте:\n",
" # text = re.sub(r'\\b\\w+\\b', '', text) # Удаляем отдельные слова (без чисел и знаков препинания)\n",
" # text = ' '.join([word for word in text.split() if word not in stop_words]) # Удаляем стоп-слова\n",
" return text\n",
"\n",
"for i, row in df.iterrows():\n",
" df.at[i, 'description'] = clean(row['description'])"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "0huKeMs4-rl_",
"outputId": "8659997c-9b8a-45bb-e2d7-fcc05422b92a"
},
"outputs": [],
"source": [
"# pip install transformers sentencepiece\n",
"\n",
"tokenizer = AutoTokenizer.from_pretrained(\"cointegrated/rubert-tiny2\")\n",
"model = AutoModel.from_pretrained(\"cointegrated/rubert-tiny2\")\n",
"# model.cuda() # uncomment it if you have a GPU"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {
"id": "Xsxq-Ohx-rmA"
},
"outputs": [],
"source": [
"# применяем токенизатор:\n",
"# -≥ add_special_tokens = добавляем служебные токены (CLS=101, EOS=102)\n",
"# -≥ truncation = обрезаем по максимальной длине\n",
"# -≥ max_length = максимальная длина последовательности\n",
"tokenized = df['description'].apply((lambda x: tokenizer.encode(x,\n",
" add_special_tokens=True,\n",
" truncation=True,\n",
" max_length=1024)))"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {
"id": "OuaXqHNj-rmB"
},
"outputs": [],
"source": [
"max_len = 1024\n",
"# Делаю пэддинг чтобы добить до max_len последовательности\n",
"padded = np.array([i + [0]*(max_len-len(i)) for i in tokenized.values])\n",
"# И маску чтобы не применять self-attention на pad\n",
"attention_mask = np.where(padded != 0, 1, 0)"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {
"id": "h3bfQh2o-rmC"
},
"outputs": [],
"source": [
"# Датасет для массивов\n",
"class BertInputs(torch.utils.data.Dataset):\n",
" def __init__(self, tokenized_inputs, attention_masks):\n",
" super().__init__()\n",
" self.tokenized_inputs = tokenized_inputs\n",
" self.attention_masks = attention_masks\n",
"\n",
" def __len__(self):\n",
" return self.tokenized_inputs.shape[0]\n",
"\n",
" def __getitem__(self, idx):\n",
" ids = self.tokenized_inputs[idx]\n",
" ams = self.attention_masks[idx]\n",
"\n",
" return ids, ams\n",
"\n",
"dataset = BertInputs(padded, attention_mask)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "Q7yYgEP3-rmC",
"outputId": "76047d40-f793-4cef-fc02-b98b232661f8"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"torch.Size([100, 1024]) torch.Size([100, 1024])\n"
]
}
],
"source": [
"#DataLoader чтобы отправлять бачи в цикл обучения\n",
"loader = torch.utils.data.DataLoader(dataset, batch_size=100, shuffle=True)\n",
"sample_ids, sample_ams = next(iter(loader))\n",
"print(sample_ids.shape, sample_ams.shape)\n",
"\n",
"# shape BATCH_SIZE x MAX_LEN - что заходит в BERT"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "r1h0BNy1-rmD",
"outputId": "adea19c9-a0f2-418c-9a21-ebe8daa00077"
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 94/94 [01:13<00:00, 1.28it/s]"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"CPU times: user 1min 10s, sys: 145 ms, total: 1min 10s\n",
"Wall time: 1min 13s\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"\n"
]
}
],
"source": [
"%%time\n",
"\n",
"vectors_in_batch = []\n",
"\n",
"# Iterate over all batches\n",
"for inputs, attention_masks in tqdm(loader):\n",
" vectors_in_mini_batch = [] # Store vectors in mini-batch\n",
" with torch.no_grad():\n",
" last_hidden_states = model(inputs.cuda(), attention_mask=attention_masks.cuda())\n",
" vector = last_hidden_states[0][:,0,:].detach().cpu().numpy()\n",
" vectors_in_mini_batch.append(vector)\n",
"\n",
" vectors_in_batch.extend(vectors_in_mini_batch)"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [],
"source": [
"import itertools\n",
"\n",
"# Open the file and load the nested list\n",
"vectors_in_batch = load('vectors_in_batch.joblib')\n",
"\n",
"# Convert the nested list to an unnested list\n",
"text_embeddings = list(itertools.chain.from_iterable(vectors_in_batch))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Сохранение эмбеддингов\n",
"dump(vectors_in_batch, 'vectors_in_batch.joblib')"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"94"
]
},
"execution_count": 17,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"len(vectors_in_batch)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"9366"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"len(text_embeddings)"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"gpuType": "T4",
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.4"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 0
}
|