Created
April 19, 2021 16:44
-
-
Save ShairozS/d79c4534ae8ca9745407ca42079f945b to your computer and use it in GitHub Desktop.
Form triplets from a batch of Pytorch tensor embeddings
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| def form_triplets(inA, inB): | |
| ''' | |
| Form triplets from two tensors of embeddings. It is assumed that the embeddings at corresponding batch positions are similar | |
| and all other batch positions are dissimilar | |
| i.e inA[i] ~ inB[i] and inA[i] !~ inB[j] for all i =! j | |
| ''' | |
| b, emb_size = inA.shape | |
| perms = b**2 | |
| labels = [0]*perms; sim_idxs = [(0 + i*b) + i for i in range(b)] | |
| for idx in sim_idxs: | |
| labels[idx] = 1 | |
| labels = torch.Tensor(labels) | |
| labels = labels.type(torch.BoolTensor).to(inA.device) | |
| anchors = inA.repeat(b, 1)[~labels] | |
| negatives = torch.cat([inB[i,:].repeat(b,1) for i in range(b)])[~labels] | |
| positives = inB.repeat(b, 1)[~labels] | |
| return(anchors, positives, negatives) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment