def _build_modules(self):
"""Defines method to build all sub-modules of the model"""
self.blocks = nn.Sequential()
for index, layer_config in enumerate(self.config):
layer = layer_factory.create(layer_config['name'],
**layer_config['params'])
self.blocks.add_module(name='{}_{}'.format(
index, layer_config['name'].lower()),
module=layer)
def test_swish(self):
"""Test swish activation function"""
layer_name = 'Swish'
args = {}
layer = layer_factory.create(layer_name, **args)
x = torch.rand(10)
out = layer(x)
target = x * torch.sigmoid(x)
assert_array_equal(target, out)
def test_flatten(self):
"""Test creation of a Flatten layer"""
layer_name = 'Flatten'
args = {}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, 16, 16))
y = layer(x)
self.assertEqual(y.shape, torch.Size([1, 256]))
def test_adaptiveavgpool2d(self):
"""Test creation of a adaptiveavgpool2d layer"""
layer_name = 'AdaptiveAvgPool2d'
args = {
'output_size': [1, 1],
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, 16, 128, 1024))
y = layer(x)
self.assertEqual(y.shape, torch.Size([1, 16, *args['output_size']]))
def test_maxpool2d(self):
"""Test creation of a maxpool2d layer"""
layer_name = 'MaxPool2d'
args = {
'kernel_size': [3, 7],
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, 16, 128, 1024))
y = layer(x)
self.assertEqual(y.shape, torch.Size([1, 16, 42, 146]))
def test_dropout(self):
"""Test creation of a dropout layer"""
layer_name = 'Dropout'
args = {
'p': 0.5,
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, 1, 128, 1024))
y = layer(x)
self.assertEqual(y.shape, x.shape)
def test_batchnorm2d(self):
"""Test creation of a batchnorm2d layer"""
layer_name = 'BatchNorm2d'
args = {
'num_features': 3,
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, args['num_features'], 128, 1024))
y = layer(x)
self.assertEqual(y.shape, x.shape)
def test_linear(self):
"""Test creation of a linear layer"""
layer_name = 'Linear'
args = {
'in_features': 16,
'out_features': 8
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros(args['in_features'])
y = layer(x)
self.assertEqual(y.shape, torch.Size([args['out_features']]))
def test_conv2d(self):
"""Test creation of a Conv2D layer"""
layer_name = 'Conv2d'
args = {
'in_channels': 1,
'out_channels': 16,
'kernel_size': [3, 7]
}
layer = layer_factory.create(layer_name, **args)
self.assertEqual(layer._get_name(), layer_name)
x = torch.zeros((1, args['in_channels'], 128, 1024))
y = layer(x)
self.assertEqual(y.shape, torch.Size([1, args['out_channels'], 126, 1018]))