LITENetwork¶
- class LITENetwork(n_filters=32, kernel_size=40, strides=1, activation='relu')[source]¶
LITE Network.
LITE deep neural network architecture from [1]_.
- Parameters:
- n_filtersint or list of int32, default = 32
The number of filters used in one lite layer, if not a list, the same number of filters is used in all lite layers.
- kernel_sizeint or list of int, default = 40
The head kernel size used for each lite layer, if not a list, the same is used in all lite layers.
- stridesint or list of int, default = 1
The strides of kernels in convolution layers for each lite layer, if not a list, the same is used in all lite layers.
- activationstr or list of str, default = ‘relu’
The activation function used in each lite layer, if not a list, the same is used in all lite layers.
Notes
..[1] Ismail-Fawaz et al. LITE: Light Inception with boosTing tEchniques for Time Series Classificaion, IEEE International Conference on Data Science and Advanced Analytics, 2023.
Adapted from the implementation from Ismail-Fawaz et. al
https://github.com/MSD-IRIMAS/LITE
Methods
build_network(input_shape, **kwargs)Construct a network and return its input and output layers.
hybrid_layer(input_tensor, input_channels[, ...])Construct the hybrid layer to compute features of custom filters.
- hybrid_layer(input_tensor, input_channels, kernel_sizes=None)[source]¶
Construct the hybrid layer to compute features of custom filters.
- Parameters:
- input_tensortensorflow tensor, usually the input layer of the model.
- input_channelsint, the number of input channels in case of multivariate.
- kernel_sizeslist of int, default = [2,4,8,16,32,64],
- the size of the hand-crafted filters.
- Returns:
- hybrid_layertensorflow tensor containing the concatenation
- of the output features extracted form hand-crafted convolution filters.