How to save and load my neural network model after training along with weights in python?

Question

I have trained a single layer neural network model in python (a simple model without keras and tensorflow). How canI save it after training along with weights in python, and how to load it later?

Answer 1

All the trainable parameters like weights and biases could be treated as either Python lists or NumPy arrays ( which is mostly preferred ).

For Python lists :

If your trainable parameters are Python lists then you can use pickle. You can pickle them like this :

import pickle
# weights is a Python array
pickle.dump( weights , open( 'weights.pkl' , 'wb' ) )

You can group together several objects in a set or list and pickle that so you have a single file. For reading it,

weights = pickle.load( open( 'weights.pkl' , 'rb' ))

For NumPy arrays :

That makes all the code easy. A NumPy array could be saved by using np.array.save() method.

np.save( 'weights.npy' , weights )

And load it,

weights = np.load( 'weights.npy' )

Apart from these prevalent methods like writing the weights and biases to a text file or a csv file may also work. Also, a JSON file may help.

Answer 2

So you write it down yourself. You need some simple steps:

1. In your code for neural network, store weights in a variable. It could be simply done by using self.weights.weights are numpy ndarrays. for example if weights are between layer with 10 neurons to layer with 100 neurons, it is a 10 * 100(or 100* 10) nd array.
2. Use numpy.save to save the ndarray.
3. For next use of your network, use numpy.load to load weights
4. In the first initialization of your network, use weights you've loaded. Don't forget, if your network is trained, weights should be frozen. It can be done by zeroing learning rate.