flexible neural network structure
This merge requests adds the following features to the CNNWhale class:
-
Number, dimensions, and properties of convolution and dense layers in CNNWhale can be specified via args to create_net_structure (neural_networks.py)
-
Possibility to apply drop-out to convolutional layers
-
Validation accuracy is printed during training (neural_networks.py)
-
train method returns average cost of the latest completed training epoch
-
train method accepts additional input args, such as batch size, number of epochs, drop-out probability, etc.
-
Option to specify verbosity level
It also makes it possible to
- generate a Morlet wavelet with frequency that changes linearly with time (in audio_signal.py)