Hi,
I am trying to compute the TripleLossFn for the given vectors, but i’m getting a loss of 0.3517 instead of 0.7. Relevant outputted matrices are given below:
As can be seen the second row in the scores matrix is [-0.9535, -1], so in this case there is no closest_negative that is less than the diagonal entry -1.
With the suggested computations the last row of the matrix negative_without_positive is [-2.953, -2] which implies -2 would incorrectly be the maximum.
Am I missing something here? This is my id if you need to look into my code: euibmtfxumzt
Thanks!