Saving and loading a trained network

Save

A network can be easily saved to a file:

from nimblenet.neuralnet import NeuralNet
# Create a network
network = NeuralNet({
    "n_inputs" : 2,
    "layers"   : [ (1, sigmoid_function) ],
})
# Save the network to disk
network.save_network_to_file( "%s.pkl" % "filename" )

In addition to doing this explicitly, all of the learning algorithms also offer the possibility to save the network after the training has completed. This is done by passing the named parameter save_trained_network = True when calling the learning function:

RMSprop( ..., save_trained_network = False ) # omitted parameters for readability

This will promt the user whether to save the network or not, upon completion of the training.

Load

If you have saved a network to a file, you can easily load the network back up by calling:

from nimblenet.neuralnet import NeuralNet
network = NeuralNet.load_network_from_file( "%s.pkl" % "filename" )