MediumNLPnlp-065

Negative Sampling Probability

Problem

Return the negative-sampling distribution over vocabulary items with the given counts: p_i = count_i ** 0.75 divided by the sum of count ** 0.75 over all items.

Return an empty list for empty input, and all zeros when every count is 0.

Hint

The 0.75 power flattens frequent words before normalizing.

Test cases

case 1

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

expected: [0.25,0.25,0.25,0.25]

case 2

in: [[4,1]]

expected: [0.7387961250362586,0.2612038749637414]

case 3

in: [[]]

expected: []

case 4

in: [[0,1]]

expected: [0,1]

case 5

in: [[9]]

expected: [1]

Starter codePython

def negative_sampling_prob(counts):
    # Your code here
    pass

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