Keras-surgeon: For pruning and Model optimization

Question

I am working on a Convolutional Neural Net built using Keras. And since the final model needs to be deployed on a processing unit with less processing capabilities, I started looking for ways to reduce the resources it'd consume. Luckily, I came across this library in Keras, Keras-surgeon, that helps in modifying trained Keras models.

I'm trying to delete channels in one of my Conv Net layers (and will proceed further on other layers). Although, documentation is very straightforward and simple, and people who have used this also praise it, when I tried to use it, it returns a weird type-error:

from kerassurgeon import Surgeon

surgeon = Surgeon(model_final)
layer_1 = model_final.layers[1] # selecting 2nd layer

surgeon.add_job('delete_channels', layer_1, channels=[4,7,3])
new_model = surgeon.operate()

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-292-77c377538a52> in <module>
      5 
      6 surgeon.add_job('delete_channels', layer_1, channels=[4,7,3])
----> 7 new_model = surgeon.operate()

~/anaconda3/envs/Learning/lib/python3.6/site-packages/kerassurgeon/surgeon.py in operate(self)
    150         for node in sorted_nodes:
    151             # Rebuild submodel up to this node
--> 152             sub_output_nodes = utils.get_node_inbound_nodes(node)
    153             outputs, output_masks = self._rebuild_graph(self.model.inputs,
    154                                                         sub_output_nodes)

~/anaconda3/envs/Learning/lib/python3.6/site-packages/kerassurgeon/utils.py in get_node_inbound_nodes(node)
     88 def get_node_inbound_nodes(node):
     89     return [get_inbound_nodes(node.inbound_layers[i])[node_index]
---> 90             for i, node_index in enumerate(node.node_indices)]
     91 
     92 

TypeError: 'int' object is not iterable

1. What exactly am I doing wrong here?

2. Also, can anyone point out anything that I could use to run the keras model on, to see its resources consumption (memory and CPU etc.)? (It'll be a big help!)

3. And what other approaches I can take to have a model that takes less resources without compromising accuracy?

Any help will be much appreciated..!

Answer 1

The definition of add_job is:

def add_job(self, job, layer, *,
            channels=None, new_layer=None, node_indices=None):
  # ...

That * in the function parameters is meant to force you to use named arguments explicitly. You must call add_job as follows:

surgeon.add_job('delete_channels', layer_1, channels=[4,7,3])

Answer 2

using tfkerassurgeon will fix the issue.

pip install tfkerassurgeon

Answer 3

I have seen this error when some function expects a list of int (or objects) and I pass a single int (or object). Maybe try surgeon.add_job('delete_channels', [layer_1], channels=[4,7,3]) instead of surgeon.add_job('delete_channels', layer_1, channels=[4,7,3])