Benutzer-Werkzeuge
ws2021:zusaetzliche_klasse_resnet
Unterschiede
Hier werden die Unterschiede zwischen zwei Versionen gezeigt.
| Nächste Überarbeitung | Vorhergehende Überarbeitung | ||
|
ws2021:zusaetzliche_klasse_resnet [2021/04/05 21:16] elena_kirschner angelegt |
ws2021:zusaetzliche_klasse_resnet [2021/04/06 16:40] (aktuell) annika_cibis |
||
|---|---|---|---|
| Zeile 1: | Zeile 1: | ||
| <code> | <code> | ||
| - | class block(nn.Module): | + | class ResNet(nn.Module): |
| - | def __init__(self, in_channels, out_channels, identity_downsample=None, stride=1): | + | def __init__(self, block, layers, image_channels): |
| - | super(block, self).__init__() | + | super(ResNet, self).__init__() |
| - | self.out_channels, self.in_channels = out_channels, in_channels | + | |
| - | self.expansion = 1 | + | self.history_loss = [] |
| - | self.conv1 = nn.Conv1d(self.in_channels, self.out_channels, kernel_size=15, stride=stride, padding=7) | + | self.history_eval = [] |
| - | self.bn1 = nn.BatchNorm1d(self.out_channels) | + | self.classific_accuracy_training = [] |
| - | self.conv2 = nn.Conv1d(self.out_channels, self.out_channels*2, kernel_size=9, stride=1, padding=4) | + | self.current_epoch = 0 |
| - | self.bn2 = nn.BatchNorm1d(self.out_channels*2) | + | |
| - | self.conv3 = nn.Conv1d(self.out_channels*2, self.out_channels*2, kernel_size=5, stride=1, padding=2) | + | self.in_channels = 12 |
| - | self.bn3 = nn.BatchNorm1d(self.out_channels*2) | + | self.conv1 = nn.Conv1d(12, 12, kernel_size=7, stride=2, padding=3) |
| - | self.relu = nn.ReLU() | + | self.bn1 = nn.BatchNorm1d(12) |
| - | self.identity_downsample = identity_downsample | + | self.relu = nn.ReLU() |
| - | self.lin1 = nn.Linear(out_channels*2*250, 800) | + | self.maxpool = nn.MaxPool1d(kernel_size=3, stride=2, padding=1) |
| - | self.lin2 = nn.Linear(800, 40) | + | |
| - | self.lin3 = nn.Linear(40, 10) | + | self.layer1 = self._make_layer(block, layers[0], out_channels=12, stride=1) |
| - | self.lin4 = nn.Linear(10, 2) | + | self.layer2 = self._make_layer(block, layers[1], out_channels=24, stride=2) |
| + | self.layer3 = self._make_layer(block, layers[2], out_channels=24*2, stride=2) | ||
| + | self.layer4 = self._make_layer(block, layers[3], out_channels=24*4, stride=2) | ||
| + | |||
| + | self.avgpool = nn.AdaptiveAvgPool1d(25) | ||
| + | self.lin1 = nn.Linear(4800, 800) | ||
| + | self.lin2 = nn.Linear(800, 40) | ||
| + | self.lin3 = nn.Linear(40, 10) | ||
| + | self.lin4 = nn.Linear(10, 2) | ||
| def forward(self, x): | def forward(self, x): | ||
| - | identity = x | ||
| x = self.conv1(x) | x = self.conv1(x) | ||
| x = self.bn1(x) | x = self.bn1(x) | ||
| x = self.relu(x) | x = self.relu(x) | ||
| - | x = self.conv2(x) | + | x = self.maxpool(x) |
| - | x = self.bn2(x) | + | |
| - | x = self.relu(x) | + | x = self.layer1(x) |
| - | x = self.conv3(x) | + | x = self.layer2(x) |
| - | x = self.bn3(x) | + | |
| - | if self.identity_downsample is not None: | + | x = self.layer3(x) |
| - | identity = self.identity_downsample(identity) | + | x = self.layer4(x) |
| - | x += identity | + | |
| + | x = self.avgpool(x) | ||
| + | x = x.reshape(x.shape[0], -1) # eventuell auf x.view Funktion umsteigen | ||
| + | |||
| + | x = F.relu(self.lin1(x)) | ||
| + | x = F.relu(self.lin2(x)) | ||
| + | x = F.relu(self.lin3(x)) | ||
| + | x = self.lin4(x) | ||
| return x | return x | ||
| + | |||
| + | def _make_layer(self, block, num_residual_blocks, out_channels, stride): | ||
| + | identity_downsample = None | ||
| + | layers = [] | ||
| + | |||
| + | if stride != 1 or self.in_channels != out_channels*2: | ||
| + | identity_downsample = nn.Sequential(nn.Conv1d(self.in_channels, out_channels*2, kernel_size=1, | ||
| + | stride=stride), | ||
| + | nn.BatchNorm1d(out_channels*2)) | ||
| + | |||
| + | layers.append(block(self.in_channels, out_channels, identity_downsample, stride=stride)) | ||
| + | |||
| + | | ||
| + | self.in_channels = out_channels*2 | ||
| + | | ||
| + | for i in range(num_residual_blocks -1): | ||
| + | layers.append(block(self.in_channels, out_channels)) | ||
| + | | ||
| + | return nn.Sequential(*layers) | ||
| + | | ||
| + | | ||
| + | def ResNet50(img_channels=12): | ||
| + | return ResNet(block, [3, 4, 6, 3], img_channels) | ||
| + | |||
| + | def test(): | ||
| + | net = ResNet50() | ||
| + | x = torch.randn(30, 12, 1000) | ||
| + | y = net(x) | ||
| + | return y | ||
| + | |||
| + | net = ResNet50() | ||
| + | learning_rate = 0.15 | ||
| + | batch_size = 50 | ||
| + | optimizer = optim.SGD(net.parameters(), lr=learning_rate) | ||
| + | |||
| </code> | </code> | ||
ws2021/zusaetzliche_klasse_resnet.1617650184.txt.gz · Zuletzt geändert: 2021/04/05 21:16 von elena_kirschner
Seiten-Werkzeuge
