I have evolved a neural network to learn y=x^2 using the neuralfit library, but I would like to save the model to do predictions later. I currently have:
import neuralfit
import numpy as np
# y(x) = x^2
x = np.arange(10).reshape(-1,1)
y = x**2
# Evolve model
model = neuralfit.Model(1,1)
model.compile('alpha', loss='mse', monitors=['size'])
model.evolve(x,y,epochs=1000)
# Save model
...
How would I save and load model?
There are two ways to do this: (1) with Neuralfit and (2) with Keras. It is best to use NeuralFit because the resulting savefile is a lot smaller (50x in this case).
Based on the documentation:
# Save a model
model.save('model.nf')
# Load a saved model
model = neuralfit.load('model.nf')
Since NeuralFit allows conversion to Keras, we can convert the model to Keras and then save it using their functionality. In other words:
# Save a model
keras_model = model.to_keras()
keras_model.save('model.h5')
# Load a saved model
keras_model = keras.models.load_model('model.h5')
model = neuralfit.from_keras(keras_model)