Spaces:
Running
Running
File size: 13,071 Bytes
6cfb4d3 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 |
"""
All neural network blocks and architectures in models/enhanced_cnn.py are custom implementations, developed to expand the model registry for advanced polymer spectral classification. While inspired by established deep learning concepts (such as residual connections, attention mechanisms, and multi-scale convolutions), they are are unique to this project and tailored for 1D spectral data.
Registry expansion: The purpose is to enrich the available models.
Literature inspiration: SE-Net, ResNet, Inception.
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
class AttentionBlock1D(nn.Module):
"""1D attention mechanism for spectral data."""
def __init__(self, channels: int, reduction: int = 8):
super().__init__()
self.channels = channels
self.global_pool = nn.AdaptiveAvgPool1d(1)
self.fc = nn.Sequential(
nn.Linear(channels, channels // reduction),
nn.ReLU(inplace=True),
nn.Linear(channels // reduction, channels),
nn.Sigmoid(),
)
def forward(self, x):
# x shape: [batch, channels, length]
b, c, _ = x.size()
# Global average pooling
y = self.global_pool(x).view(b, c)
# Fully connected layers
y = self.fc(y).view(b, c, 1)
# Apply attention weights
return x * y.expand_as(x)
class EnhancedResidualBlock1D(nn.Module):
"""Enhanced residual block with attention and improved normalization."""
def __init__(
self,
in_channels: int,
out_channels: int,
kernel_size: int = 3,
use_attention: bool = True,
dropout_rate: float = 0.1,
):
super().__init__()
padding = kernel_size // 2
self.conv1 = nn.Conv1d(in_channels, out_channels, kernel_size, padding=padding)
self.bn1 = nn.BatchNorm1d(out_channels)
self.relu = nn.ReLU(inplace=True)
self.conv2 = nn.Conv1d(out_channels, out_channels, kernel_size, padding=padding)
self.bn2 = nn.BatchNorm1d(out_channels)
self.dropout = nn.Dropout1d(dropout_rate) if dropout_rate > 0 else nn.Identity()
# Skip connection
self.skip = (
nn.Identity()
if in_channels == out_channels
else nn.Sequential(
nn.Conv1d(in_channels, out_channels, kernel_size=1),
nn.BatchNorm1d(out_channels),
)
)
# Attention mechanism
self.attention = (
AttentionBlock1D(out_channels) if use_attention else nn.Identity()
)
def forward(self, x):
identity = self.skip(x)
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.dropout(out)
out = self.conv2(out)
out = self.bn2(out)
# Apply attention
out = self.attention(out)
out = out + identity
return self.relu(out)
class MultiScaleConvBlock(nn.Module):
"""Multi-scale convolution block for capturing features at different scales."""
def __init__(self, in_channels: int, out_channels: int):
super().__init__()
# Different kernel sizes for multi-scale feature extraction
self.conv1 = nn.Conv1d(in_channels, out_channels // 4, kernel_size=3, padding=1)
self.conv2 = nn.Conv1d(in_channels, out_channels // 4, kernel_size=5, padding=2)
self.conv3 = nn.Conv1d(in_channels, out_channels // 4, kernel_size=7, padding=3)
self.conv4 = nn.Conv1d(in_channels, out_channels // 4, kernel_size=9, padding=4)
self.bn = nn.BatchNorm1d(out_channels)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
# Parallel convolutions with different kernel sizes
out1 = self.conv1(x)
out2 = self.conv2(x)
out3 = self.conv3(x)
out4 = self.conv4(x)
# Concatenate along channel dimension
out = torch.cat([out1, out2, out3, out4], dim=1)
out = self.bn(out)
return self.relu(out)
class EnhancedCNN(nn.Module):
"""Enhanced CNN with attention, multi-scale features, and improved architecture."""
def __init__(
self,
input_length: int = 500,
num_classes: int = 2,
dropout_rate: float = 0.2,
use_attention: bool = True,
):
super().__init__()
self.input_length = input_length
self.num_classes = num_classes
# Initial feature extraction
self.initial_conv = nn.Sequential(
nn.Conv1d(1, 32, kernel_size=7, padding=3),
nn.BatchNorm1d(32),
nn.ReLU(inplace=True),
nn.MaxPool1d(kernel_size=2),
)
# Multi-scale feature extraction
self.multiscale_block = MultiScaleConvBlock(32, 64)
self.pool1 = nn.MaxPool1d(kernel_size=2)
# Enhanced residual blocks
self.res_block1 = EnhancedResidualBlock1D(64, 96, use_attention=use_attention)
self.pool2 = nn.MaxPool1d(kernel_size=2)
self.res_block2 = EnhancedResidualBlock1D(96, 128, use_attention=use_attention)
self.pool3 = nn.MaxPool1d(kernel_size=2)
self.res_block3 = EnhancedResidualBlock1D(128, 160, use_attention=use_attention)
# Global feature extraction
self.global_pool = nn.AdaptiveAvgPool1d(1)
# Calculate feature size after convolutions
self.feature_size = 160
# Enhanced classifier with dropout
self.classifier = nn.Sequential(
nn.Linear(self.feature_size, 256),
nn.BatchNorm1d(256),
nn.ReLU(inplace=True),
nn.Dropout(dropout_rate),
nn.Linear(256, 128),
nn.BatchNorm1d(128),
nn.ReLU(inplace=True),
nn.Dropout(dropout_rate),
nn.Linear(128, 64),
nn.BatchNorm1d(64),
nn.ReLU(inplace=True),
nn.Dropout(dropout_rate / 2),
nn.Linear(64, num_classes),
)
# Initialize weights
self._initialize_weights()
def _initialize_weights(self):
"""Initialize model weights using Xavier initialization."""
for m in self.modules():
if isinstance(m, nn.Conv1d):
nn.init.xavier_uniform_(m.weight)
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.xavier_uniform_(m.weight)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm1d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def forward(self, x):
# Ensure input is 3D: [batch, channels, length]
if x.dim() == 2:
x = x.unsqueeze(1)
# Feature extraction
x = self.initial_conv(x)
x = self.multiscale_block(x)
x = self.pool1(x)
x = self.res_block1(x)
x = self.pool2(x)
x = self.res_block2(x)
x = self.pool3(x)
x = self.res_block3(x)
# Global pooling
x = self.global_pool(x)
x = x.view(x.size(0), -1)
# Classification
x = self.classifier(x)
return x
def get_feature_maps(self, x):
"""Extract intermediate feature maps for visualization."""
if x.dim() == 2:
x = x.unsqueeze(1)
features = {}
x = self.initial_conv(x)
features["initial"] = x
x = self.multiscale_block(x)
features["multiscale"] = x
x = self.pool1(x)
x = self.res_block1(x)
features["res1"] = x
x = self.pool2(x)
x = self.res_block2(x)
features["res2"] = x
x = self.pool3(x)
x = self.res_block3(x)
features["res3"] = x
return features
class EfficientSpectralCNN(nn.Module):
"""Efficient CNN designed for real-time inference with good performance."""
def __init__(self, input_length: int = 500, num_classes: int = 2):
super().__init__()
# Efficient feature extraction with depthwise separable convolutions
self.features = nn.Sequential(
# Initial convolution
nn.Conv1d(1, 32, kernel_size=7, padding=3),
nn.BatchNorm1d(32),
nn.ReLU(inplace=True),
nn.MaxPool1d(2),
# Depthwise separable convolutions
self._make_depthwise_sep_conv(32, 64),
nn.MaxPool1d(2),
self._make_depthwise_sep_conv(64, 96),
nn.MaxPool1d(2),
self._make_depthwise_sep_conv(96, 128),
nn.MaxPool1d(2),
# Final feature extraction
nn.Conv1d(128, 160, kernel_size=3, padding=1),
nn.BatchNorm1d(160),
nn.ReLU(inplace=True),
nn.AdaptiveAvgPool1d(1),
)
# Lightweight classifier
self.classifier = nn.Sequential(
nn.Linear(160, 64),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
nn.Linear(64, num_classes),
)
self._initialize_weights()
def _make_depthwise_sep_conv(self, in_channels, out_channels):
"""Create depthwise separable convolution block."""
return nn.Sequential(
# Depthwise convolution
nn.Conv1d(
in_channels, in_channels, kernel_size=3, padding=1, groups=in_channels
),
nn.BatchNorm1d(in_channels),
nn.ReLU(inplace=True),
# Pointwise convolution
nn.Conv1d(in_channels, out_channels, kernel_size=1),
nn.BatchNorm1d(out_channels),
nn.ReLU(inplace=True),
)
def _initialize_weights(self):
"""Initialize model weights."""
for m in self.modules():
if isinstance(m, nn.Conv1d):
nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.xavier_uniform_(m.weight)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm1d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def forward(self, x):
if x.dim() == 2:
x = x.unsqueeze(1)
x = self.features(x)
x = x.view(x.size(0), -1)
x = self.classifier(x)
return x
class HybridSpectralNet(nn.Module):
"""Hybrid network combining CNN and attention mechanisms."""
def __init__(self, input_length: int = 500, num_classes: int = 2):
super().__init__()
# CNN backbone
self.cnn_backbone = nn.Sequential(
nn.Conv1d(1, 64, kernel_size=7, padding=3),
nn.BatchNorm1d(64),
nn.ReLU(inplace=True),
nn.MaxPool1d(2),
nn.Conv1d(64, 128, kernel_size=5, padding=2),
nn.BatchNorm1d(128),
nn.ReLU(inplace=True),
nn.MaxPool1d(2),
nn.Conv1d(128, 256, kernel_size=3, padding=1),
nn.BatchNorm1d(256),
nn.ReLU(inplace=True),
)
# Self-attention layer
self.attention = nn.MultiheadAttention(
embed_dim=256, num_heads=8, dropout=0.1, batch_first=True
)
# Final pooling and classification
self.global_pool = nn.AdaptiveAvgPool1d(1)
self.classifier = nn.Sequential(
nn.Linear(256, 128),
nn.ReLU(inplace=True),
nn.Dropout(0.2),
nn.Linear(128, num_classes),
)
def forward(self, x):
if x.dim() == 2:
x = x.unsqueeze(1)
# CNN feature extraction
x = self.cnn_backbone(x)
# Prepare for attention: [batch, length, channels]
x = x.transpose(1, 2)
# Self-attention
attn_out, _ = self.attention(x, x, x)
# Back to [batch, channels, length]
x = attn_out.transpose(1, 2)
# Global pooling and classification
x = self.global_pool(x)
x = x.view(x.size(0), -1)
x = self.classifier(x)
return x
def create_enhanced_model(model_type: str = "enhanced", **kwargs):
"""Factory function to create enhanced models."""
models = {
"enhanced": EnhancedCNN,
"efficient": EfficientSpectralCNN,
"hybrid": HybridSpectralNet,
}
if model_type not in models:
raise ValueError(
f"Unknown model type: {model_type}. Available: {list(models.keys())}"
)
return models[model_type](**kwargs)
|