Initialization failure

[dalinzhangzdl]

D:\jupyter notebook\lsuv_init.py in LSUVinit(model, batch, verbose, margin, max_iter)
64 weights_and_biases = layer.get_weights()
65 weights_and_biases[0] /= np.sqrt(variance) / np.sqrt(needed_variance)
---> 66 layer.set_weights(weights_and_biases)
67 weights /= np.sqrt(variance) / np.sqrt(needed_variance)
68 layer.set_weights([weights, biases])

UnboundLocalError: local variable 'weights' referenced before assignment

[shaginhekvs]

It should be changed to following :

layer.set_weights(weights_and_biases)
weights_and_biases[1] /= np.sqrt(variance) / np.sqrt(needed_variance)
layer.set_weights(weights_and_biases)

but [1] actually has biases of the layer, and according to the paper, we're supposed to just scale the weights not the biases, so why is the code initializing biases as well ?

[andrasmolnar]

Just remove lines 66 and 67 to make it work again:

            weights /= np.sqrt(variance) / np.sqrt(needed_variance)
            layer.set_weights([weights, biases])