Mask the top k elements in a tensor in PyTorch (different k for each row)

user_13043216 0
Viewed 126

For any 2D tensor X, how to get the mask for top K elements for each row where K is a tensor (not restricted to an int)?

Input:

tensor([[0.6607, 0.1165, 0.0278, 0.1950],
        [0.0529, 0.4607, 0.2729, 0.2135],
        [0.3267, 0.0902, 0.4578, 0.1253]])

Desired output: for K = torch.tensor([2,3,1])

tensor([[ True,  False, False, True],
        [ False, True,  True,  True],
        [ False, False, True, False]])

I have tried these [1], [2], but can not succeed.

pytorch
1 Answers

You can use the torch.topk and torch.tensor.scatter_ methods for this:

K = torch.tensor([2,3,1])
for idx, k in enumerate(K):
    top_k = torch.topk(x[idx], k)
    x[idx].scatter_(0, top_k.indices, 1)
mask = x.eq(1)
user_170794640
Related