Spaces:
Sleeping
Sleeping
File size: 12,493 Bytes
68545bc |
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 |
"""
Нейросетевые методы классификации текстов: MLP, CNN, LSTM, GRU, гибридные архитектуры.
Примечание: Для трансформеров (BERT, RuBERT) требуется установка transformers и torch.
"""
from __future__ import annotations
import time
from dataclasses import dataclass
from typing import List, Dict, Any, Optional, Tuple
import numpy as np
import pandas as pd
try:
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, models, callbacks
TENSORFLOW_AVAILABLE = True
except ImportError:
TENSORFLOW_AVAILABLE = False
print("⚠️ TensorFlow не установлен. Нейросетевые модели недоступны.")
try:
import torch
import torch.nn as nn
from transformers import AutoTokenizer, AutoModel
TRANSFORMERS_AVAILABLE = True
except ImportError:
TRANSFORMERS_AVAILABLE = False
print("⚠️ PyTorch/Transformers не установлены. Трансформерные модели недоступны.")
@dataclass
class NeuralConfig:
"""Конфигурация нейросетевой модели."""
model_type: str # mlp, cnn, lstm, gru, cnn_lstm, birnn_attention
input_dim: int
num_classes: int
embedding_dim: int = 300
hidden_dim: int = 128
dropout: float = 0.5
learning_rate: float = 0.001
epochs: int = 10
batch_size: int = 32
validation_split: float = 0.2
class NeuralClassifiers:
"""Класс для работы с нейросетевыми классификаторами."""
def __init__(self, config: NeuralConfig):
if not TENSORFLOW_AVAILABLE:
raise ImportError("TensorFlow не установлен. Установите: pip install tensorflow")
self.config = config
self.model = self._create_model()
self.history = None
self.train_time = 0.0
self.predict_time = 0.0
def _create_model(self):
"""Создает нейросетевую модель."""
model_type = self.config.model_type.lower()
if model_type == "mlp":
return self._create_mlp()
elif model_type == "cnn":
return self._create_cnn()
elif model_type == "lstm":
return self._create_lstm()
elif model_type == "gru":
return self._create_gru()
elif model_type == "cnn_lstm":
return self._create_cnn_lstm()
elif model_type == "birnn_attention":
return self._create_birnn_attention()
else:
raise ValueError(f"Неизвестный тип модели: {model_type}")
def _create_mlp(self):
"""Многослойный персептрон."""
model = models.Sequential([
layers.Dense(self.config.hidden_dim, activation='relu', input_dim=self.config.input_dim),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.hidden_dim // 2, activation='relu'),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def _create_cnn(self):
"""Сверточная нейросеть для текста (Kim CNN)."""
# Для CNN нужна последовательность токенов, поэтому используем embedding
# В упрощенной версии работаем с уже векторизованными данными
model = models.Sequential([
layers.Reshape((self.config.input_dim, 1), input_shape=(self.config.input_dim,)),
layers.Conv1D(128, 3, activation='relu'),
layers.MaxPooling1D(2),
layers.Conv1D(64, 3, activation='relu'),
layers.GlobalMaxPooling1D(),
layers.Dense(self.config.hidden_dim, activation='relu'),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def _create_lstm(self):
"""LSTM сеть."""
model = models.Sequential([
layers.Reshape((self.config.input_dim, 1), input_shape=(self.config.input_dim,)),
layers.LSTM(self.config.hidden_dim, return_sequences=False),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def _create_gru(self):
"""GRU сеть."""
model = models.Sequential([
layers.Reshape((self.config.input_dim, 1), input_shape=(self.config.input_dim,)),
layers.GRU(self.config.hidden_dim, return_sequences=False),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def _create_cnn_lstm(self):
"""Гибридная CNN + LSTM архитектура."""
model = models.Sequential([
layers.Reshape((self.config.input_dim, 1), input_shape=(self.config.input_dim,)),
layers.Conv1D(64, 3, activation='relu'),
layers.MaxPooling1D(2),
layers.LSTM(self.config.hidden_dim, return_sequences=False),
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def _create_birnn_attention(self):
"""Двунаправленная RNN с механизмом внимания (упрощенная версия)."""
# Упрощенная версия без настоящего attention механизма
model = models.Sequential([
layers.Reshape((self.config.input_dim, 1), input_shape=(self.config.input_dim,)),
layers.Bidirectional(layers.LSTM(self.config.hidden_dim, return_sequences=True)),
layers.GlobalAveragePooling1D(), # Простая агрегация вместо attention
layers.Dropout(self.config.dropout),
layers.Dense(self.config.num_classes, activation='softmax')
])
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=self.config.learning_rate),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model
def fit(self, X, y, validation_data=None):
"""Обучение модели."""
if not TENSORFLOW_AVAILABLE:
raise ImportError("TensorFlow не установлен")
start = time.time()
callbacks_list = [
callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True),
callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=2, min_lr=1e-7)
]
if validation_data is None and self.config.validation_split > 0:
self.history = self.model.fit(
X, y,
epochs=self.config.epochs,
batch_size=self.config.batch_size,
validation_split=self.config.validation_split,
callbacks=callbacks_list,
verbose=1
)
else:
self.history = self.model.fit(
X, y,
epochs=self.config.epochs,
batch_size=self.config.batch_size,
validation_data=validation_data,
callbacks=callbacks_list,
verbose=1
)
self.train_time = time.time() - start
return self
def predict(self, X):
"""Предсказание классов."""
start = time.time()
predictions = self.model.predict(X, verbose=0)
self.predict_time = time.time() - start
return np.argmax(predictions, axis=1)
def predict_proba(self, X):
"""Предсказание вероятностей."""
return self.model.predict(X, verbose=0)
class TransformerClassifier:
"""
Классификатор на основе трансформеров (BERT, RuBERT).
Требует установки transformers и torch.
"""
def __init__(self, model_name: str = "DeepPavlov/rubert-base-cased",
num_classes: int = 2,
max_length: int = 512,
learning_rate: float = 2e-5,
epochs: int = 3,
batch_size: int = 16):
if not TRANSFORMERS_AVAILABLE:
raise ImportError(
"PyTorch и Transformers не установлены. "
"Установите: pip install torch transformers"
)
self.model_name = model_name
self.num_classes = num_classes
self.max_length = max_length
self.learning_rate = learning_rate
self.epochs = epochs
self.batch_size = batch_size
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
self.model = AutoModel.from_pretrained(model_name)
# Добавляем классификационный слой
self.classifier = nn.Sequential(
nn.Linear(self.model.config.hidden_size, 256),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(256, num_classes)
)
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.model.to(self.device)
self.classifier.to(self.device)
def fit(self, texts: List[str], labels: List[int]):
"""Обучение трансформерной модели."""
# Реализация обучения требует более сложной логики
# Здесь только заглушка
raise NotImplementedError(
"Полная реализация обучения трансформеров требует дополнительной настройки. "
"Рекомендуется использовать готовые решения из библиотеки transformers."
)
def predict(self, texts: List[str]):
"""Предсказание классов."""
raise NotImplementedError("См. fit()")
if __name__ == "__main__":
# Тестирование (только если TensorFlow доступен)
if TENSORFLOW_AVAILABLE:
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
X, y = make_classification(n_samples=1000, n_features=100, n_classes=3, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
config = NeuralConfig(
model_type="mlp",
input_dim=100,
num_classes=3,
epochs=5
)
classifier = NeuralClassifiers(config)
classifier.fit(X_train, y_train)
predictions = classifier.predict(X_test)
from sklearn.metrics import accuracy_score
print(f"Точность: {accuracy_score(y_test, predictions):.4f}")
else:
print("TensorFlow не установлен. Тесты пропущены.")
|