Flattening and Unflattening Keras Model Weights

I am implementing an Evolutionary Strategies(ES) approach to Reinforcement Learning benchmark problems from Gym as documented here. In ES you optimize parameters theta as a black box. Thus instead of training our keras model with backpropagation we are directly optimizing the parameter space via random perterbations (mutations) in a population, and then selection. Thus we would like to extract model weights to be manipulated directly, hence the need for flattening and unflattening of the weights, which can be done with the following code:

import collections
def flatten(weights):
    w = []
    for l in weights:
	if isinstance(l, collections.Iterable):
	    w = w + flatten(l)
	    w = w + [l]
    return w

shape = [2,3]
def unflatten(weights, shape, model):
    w = []
    i = 0
    for l, size in enumerate(shape):
	layer = model.layers[l].get_weights()
	params = layer[0]
	bias = layer[1]
	new_layer = []
	new_params = []
	new_bias = []
	for param in params:
	    i += size
	for b in bias:
	    i += 1
    return w

It took me an hour or so to get these right, so I am documenting them here incase I need to reference them again in the future.

Written on March 21, 2018