Contrastive Graph Loss Lite

Problem

Compute a contrastive graph loss over aligned positive and negative pairs: -log(sigmoid((pos - neg) / temperature)).

pos_scores and neg_scores are equal-length lists of similarity scores. Return the mean loss over pairs, or 0.0 when the lists are empty.

Hint

The loss shrinks as positive scores pull ahead of negative scores.

case 1

in: [[0.9,0.7],[0.1,0.3],1]

expected: 0.44205795917386514

case 2

in: [[0.5],[0.5],2]

expected: 0.6931471805599453

case 3

in: [[],[],1]

expected: 0

Starter codePython

import math
def contrastive_graph_loss(pos_scores, neg_scores, temperature):
    # Your code here
    pass

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