1- Weight pruning which focuses on individual weights and set them to zero if they are less than a certain threshold value.
2- Neuron pruning which set whole columns to zero according to their L2 Norm values
Just run the notebook in colab and every thing will run smoothly.
In the Tensorflow_Pruning_of_Weights.ipynb , this notebook is with biases.
In this piece of code , the step is "2" because odd values in the array of trainable_weights stands for the biases
for i in range(0,len(model.trainable_weights),2):
Weights.append(model.trainable_weights[i].numpy().reshape(-1))
There is a slight performance difference between with biases and no biases. The no biases perfomance decrease slower than with biases as i believe biases kind of compensate for the weights lost.