For simple convolution neural network (without residual, dense connections) You can specify architecture in yaml file.
Example of yaml file:
version: 1
defaults:
experiment_name: "baseline"
epochs: 5
batch_size: 64
optimizer:
name: "Adam"
learning_rate: 0.001
beta1: 0.9
beta2: 0.999
architecture:
convolution_layers:
conv_layer_1:
type: "Conv2d"
in_channels: 3
out_channels: 32
kernel_size: (3,3)
stride: (1,1)
padding: (1,1)
pooling_1:
type: "MaxPool2d"
kernel_size: (2,2)
stride: (1,1)
activation_function_1:
type: "ReLU"
conv_layer_2:
type: "Conv2d"
in_channels: 32
out_channels: 64
kernel_size: (3,3)
stride: (1,1)
padding: (1,1)
pooling_2:
type: "MaxPool2d"
kernel_size: (2,2)
stride: (1,1)
activation_function_2:
type: "ReLU"
conv_layer_3:
type: "Conv2d"
in_channels: 64
out_channels: 128
kernel_size: (3,3)
stride: (1,1)
padding: (1,1)
pooling_3:
type: "MaxPool2d"
kernel_size: (2,2)
activation_function_3:
type: "ReLU"
fully_connected_layers:
fc_layer_1:
type: "Linear"
in_features: 2048
out_features: 10
When yaml is finished You just need to run:
python main.py --config my_yaml_file.yamlFor more complex architectures you can create class which name need to be CustomCnnModelCifar10 where You will specify architecture. When class is created with architecture You need to create yaml file.
version: 1
defaults:
experiment_name: "deepmodel"
epochs: 20
batch_size: 64
optimizer:
name: "Adam"
learning_rate: 0.001
beta1: 0.9
beta2: 0.999
architecture:
class: "CustomCnnModelCifar10"Where you will specify class.