Closed james34602 closed 5 years ago
building a NN without extra overhead
What overhead? It is a bad idea to print floating point numbers in plain text. KANN provides functions to save/load models and parameters. Please use those.
What my thoughts is user have chance to explore their model easier, or I would say print the coefficients with just one function call. Something like get_weights() in Keras.
Nothing bad to debug their model via weights monitoring.
I would be in favor of the addition of this feature, if possible. It would be nice to be able to make networks 'exportable' in some format other than the kann format, so that small networks can be put directly into code, or for other programs that support GPU training.
If a function that can print MLP to weights and bias in text file would be great. As the function could help building a NN without extra overhead.