Structured Pruning Channel Selection

Problem

Select the channels to keep when pruning by filter norm.

Keep the indices of the `keep` largest channel norms, returning them sorted ascending. Ties keep the smaller index. Return [] when keep is 0.

case 1

in: [[0.5,2,1,3],2]

expected: [1,3]

case 2

in: [[0.5,2,1,3],0]

expected: []

case 3

in: [[0.5,2,1,3],4]

expected: [0,1,2,3]

case 4

in: [[1,1,2],2]

expected: [0,2]

Starter codePython

def prune_channels(channel_norms, keep):
    # Your code here
    pass

4 test cases · run your code against them in the editor