KernelSHAP Weight

Problem

Compute the KernelSHAP coalition weight for a model with num_features features and a coalition of subset_size: (M - 1) / (C(M, z) * z * (M - z)).

Return 0.0 for endpoint coalitions (z = 0 or z = M) or when M < 2.

Hint

The Shapley kernel assigns the largest weight to small and large coalitions.

case 1

in: [4,1]

expected: 0.25

case 2

in: [4,2]

expected: 0.125

case 3

in: [5,2]

expected: 0.06666666666666667

case 4

in: [4,0]

expected: 0

Starter codePython

def kernelshap_weight(num_features, subset_size):
    # Your code here
    pass

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