Hello,

I’m curious if its possible to do something like `np.random.choice([a, b], p=[p_a, p_b])`

using PyTorch where `a`

and `b`

are 1-d tensors of length L and `p_a`

and `p_b`

are probabilities used to sample elements from either tensor `a`

or `b`

into the resultant 1-d tensor.

I think `torch.multinomial([p_a, p_b], L)`

might be useful here - it returns a 1-d tensor length `L`

of 0s and 1s based on the probabilities I give it but I’m drawing a blank of how to utilize this tensor to sample from my tensors `a`

and `b`

. Appreciate any help!