I am trying to implement the triplet loss with random negative triplet selection. Right now I have a tensor of shape (batch_size, batch_size, batch_size) where element (i,j,