Cloning of model

[tambetm]

In my application I need to make copy of the model periodically. Currently I'm saving the weights to a file and load them into another model. That's clearly inefficient.

I would allow loading of weights directly from dictionary, instead of from file. That would also solve slight disparity with between two methods: Model.serialize() returns dictionary, but Model.load_weights() expects file name. I would allow two additional methods: Model.unserialize(dict) and Model.save_weights(weights_path).