EncoderNetwork¶
- class EncoderNetwork(kernel_size=None, n_filters=None, dropout_proba=0.2, max_pool_size=2, activation='sigmoid', padding='same', strides=1, fc_units=256)[source]¶
Establish the network structure for an Encoder.
Adapted from the implementation used in [1]
- Parameters:
- kernel_sizearray of int, default = [5, 11, 21]
Specifies the length of the 1D convolution windows.
- n_filtersarray of int, default = [128, 256, 512]
Specifying the number of 1D convolution filters used for each layer, the shape of this array should be the same as kernel_size.
- max_pool_sizeint, default = 2
Size of the max pooling windows.
- activationstring, default = sigmoid
Keras activation function.
- dropout_probafloat, default = 0.2
specifying the dropout layer probability.
- paddingstring, default = “same”
Specifying the type of padding used for the 1D convolution.
- stridesint, default = 1
Specifying the sliding rate of the 1D convolution filter.
- fc_unitsint, default = 256
Specifying the number of units in the hiddent fully connected layer used in the EncoderNetwork.
Notes
Adapted from source code https://github.com/hfawaz/dl-4-tsc/blob/master/classifiers/encoder.py
References
[1]Serrà et al. Towards a Universal Neural Network Encoder for Time Series
In proceedings International Conference of the Catalan Association for Artificial Intelligence, 120–129 2018.
Methods
build_network(input_shape, **kwargs)Construct a network and return its input and output layers.