crested.tl.zoo.dilated_cnn_decoupled

crested.tl.zoo.dilated_cnn_decoupled#

crested.tl.zoo.dilated_cnn_decoupled(seq_len, num_classes, first_conv_filters=512, first_conv_filter_size=5, first_conv_pool_size=0, first_conv_activation='gelu', first_conv_l2=1e-05, first_conv_dropout=0.1, n_dil_layers=8, num_filters=512, filter_size=3, activation='relu', output_activation='softplus', l2=1e-05, dropout=0.1, batch_norm=True, dense_bias=True)#

Construct a CNN using dilated convolutions with a separate dense head per output class.

Parameters:

  • seq_len (int) – Width of the input region.

  • num_classes (int) – Number of classes to predict.

  • first_conv_filters (int (default: 512)) – Number of filters in the first convolutional layer.

  • first_conv_filter_size (int (default: 5)) – Size of the kernel in the first convolutional layer.

  • first_conv_pool_size (int (default: 0)) – Size of the pooling kernel in the first convolutional layer.

  • first_conv_activation (str (default: 'gelu')) – Activation function in the first convolutional layer.

  • first_conv_l2 (float (default: 1e-05)) – L2 regularization for the first convolutional layer.

  • first_conv_dropout (float (default: 0.1)) – Dropout rate for the first convolutional layer.

  • n_dil_layers (int (default: 8)) – Number of dilated convolutional layers.

  • num_filters (int (default: 512)) – Number of filters in the dilated convolutional layers.

  • filter_size (int (default: 3)) – Size of the kernel in the dilated convolutional layers.

  • activation (str (default: 'relu')) – Activation function in the dilated convolutional layers.

  • output_activation (str (default: 'softplus')) – Activation function for the output layer.

  • l2 (float (default: 1e-05)) – L2 regularization for the dilated convolutional layers.

  • dropout (float (default: 0.1)) – Dropout rate for the dilated convolutional layers.

  • batch_norm (bool (default: True)) – Whether or not to use batch normalization.

  • dense_bias (bool (default: True)) – Whether or not to add a bias to the dense layer.

Return type:

Model

Returns:

A Keras model.

Contents