Data Mixing Weight Effect

Problem

Return the change in loss when mixing weights are sharpened by a temperature: L(mix) - L(base), where L(mix) uses weights normalized after raising them to the temperature and L(base) uses the raw normalized weights. Return 0.0 when the inputs are empty or misaligned, temperature <= 0, or the total weight is zero.

Test cases

case 1

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

expected: 0

case 2

in: [[1,10],[0.9,0.1],2]

expected: -0.7902439024390244

case 3

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

expected: 0

case 4

in: [[],[],2]

expected: 0

Starter codePython

def mix_loss_effect(losses, weights, temperature):
    # Your code here
    pass

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