def build(self,
classifier_type='convnet',
classifier_args=dict(dropout=0.2)):
'''Builds a simple image classifier.
Args:
classifier_type (str): Network type for the classifier.
classifier_args: Classifier arguments. Can include dropout,
batch_norm, layer_norm, etc.
'''
classifier_args = classifier_args or {}
shape = self.get_dims('x', 'y', 'c')
dim_l = self.get_dims('labels')
Encoder, args = update_encoder_args(shape,
model_type=classifier_type,
encoder_args=classifier_args)
args.update(**classifier_args)
classifier = Encoder(shape, dim_out=dim_l, **args)
self.nets.classifier = classifier
def build(self,
classifier_type='convnet',
classifier_args=dict(dropout=0.2)):
classifier_args = classifier_args or {}
shape = self.get_dims('x', 'y', 'c')
dim_l = self.get_dims('labels')
Encoder, args = update_encoder_args(
shape, model_type=classifier_type, encoder_args=classifier_args)
args.update(**classifier_args)
classifier = Encoder(shape, dim_out=dim_l, **args)
self.nets.classifier = classifier
def build(self,
dim_out=None,
encoder_type: str = 'convnet',
encoder_args=dict(fully_connected_layers=1028),
Encoder=None):
x_shape = self.get_dims('x', 'y', 'c')
Encoder_, encoder_args = update_encoder_args(x_shape,
model_type=encoder_type,
encoder_args=encoder_args)
Encoder = Encoder or Encoder_
encoder = Encoder(x_shape, dim_out=dim_out, **encoder_args)
self.nets.encoder = encoder
def build(self,
classifier_type='convnet',
classifier_args=dict(dropout=0.2)):
classifier_args = classifier_args or {}
shape = self.get_dims('x', 'y', 'c')
dim_l = self.get_dims('labels')
Encoder, args = update_encoder_args(shape,
model_type=classifier_type,
encoder_args=classifier_args)
args.update(**classifier_args)
classifier = Encoder(shape, dim_out=dim_l, **args)
self.nets.classifier = classifier
def build(self,
encoder_type: str = 'convnet',
dim_out: int = None,
encoder_args=dict(fully_connected_layers=1024)):
'''
Args:
encoder_type: Encoder model type.
dim_out: Output size.
encoder_args: Arguments for encoder build.
'''
x_shape = self.get_dims('x', 'y', 'c')
Encoder, encoder_args = update_encoder_args(x_shape,
model_type=encoder_type,
encoder_args=encoder_args)
encoder = Encoder(x_shape, dim_out=dim_out, **encoder_args)
self.nets.encoder = encoder
def build(self, classifier_type='convnet',
classifier_args=dict(dropout=0.2)):
'''Builds a simple image classifier.
Args:
classifier_type (str): Network type for the classifier.
classifier_args: Classifier arguments. Can include dropout,
batch_norm, layer_norm, etc.
'''
classifier_args = classifier_args or {}
shape = self.get_dims('x', 'y', 'c')
dim_l = self.get_dims('labels')
Encoder, args = update_encoder_args(
shape, model_type=classifier_type, encoder_args=classifier_args)
args.update(**classifier_args)
classifier = Encoder(shape, dim_out=dim_l, **args)
self.nets.classifier = classifier
def build(self, encoder_type: str='convnet', dim_out: int=None,
encoder_args=dict(), semi_supervised=None):
'''
Args:
encoder_type: Encoder model type.
dim_out: Output size.
encoder_args: Arguments for encoder build.
'''
x_shape = self.get_dims('x', 'y', 'c')
if 'encoder' not in self.nets:
if encoder_type == 'alexnet':
Encoder = AlexNetEncoder
encoder_args.pop('min_dim')
encoder = Encoder(x_shape, dim_out=dim_out, **encoder_args)
else:
Encoder, _ = update_encoder_args(
x_shape, model_type=encoder_type)
encoder = Encoder(x_shape, dim_out=dim_out, **encoder_args)
self.nets.encoder = encoder
self.data.reset(mode='test', make_pbar=False)
self.data.next()
X = self.inputs('inputs').cpu()
Z = self.nets.encoder(X)
dim_z = Z.size(1)
self.linear_indices = []
self.conv_indices = []
self.c_idx = None
index = -2
try:
while True:
layer = self.nets.encoder.models[index]
if isinstance(layer, torch.nn.modules.linear.Linear):
# Is a linear layer.
self.linear_indices.append(index)
elif isinstance(layer, cortex.built_ins.networks.modules.View):
# Is a flattened version of a convolutional layer.
self.c_idx = self.c_idx or index
elif isinstance(layer, torch.nn.modules.conv.Conv2d):
# Is a convolutional layer.
self.conv_indices.append(index)
index -= 1
except IndexError:
pass
if not semi_supervised:
# Build the classifier on top of Y.
self.classifier.build(dim_in=dim_z)
# Build the classifier on top of the layer below Y.
self.h_idx = self.linear_indices[0]
dim_h = self.nets.encoder.states[self.h_idx].size(1)
self.classifier_h.build(dim_in=dim_h)
# Build the classifier on top of the last conv layer.
dim_c = self.nets.encoder.states[self.c_idx].size(1)
self.classifier_c.build(dim_in=dim_c)