How can I convert an array containing arrays of different lengths into a pytorch tensor? e.g.
devX = array([
array([1,2,3,4]),
array([1,2,3])
array([1,2,3,4,5])
])
Asked
Active
Viewed 406 times
1
-
This [answer](https://stackoverflow.com/a/35751427/6253220) might help you – Ic3fr0g Jan 31 '19 at 07:24
-
What result do you want? – hpaulj Jan 31 '19 at 08:01
-
Is it possible to do so without padding? Result should be the same as above but in a tensor – Roy Jan 31 '19 at 14:49
-
A tensor is a multi-dimensional matrix containing elements of a single data type, so I don't think you can do that. – Panos Antoniadis Jan 31 '19 at 15:43
-
Except padding you can use `PackedSequence` but would have to sort them beforehand (and keep unsorting array), not sure what's your use case though. Otherwise padding is the only way to go – Szymon Maszke Jan 31 '19 at 15:53
-
Rows must have the same sizes. `[torch.tensor(a) for a in devX]` you can use list of tensors, if its good for you – Kornel Dylski Feb 01 '19 at 13:34