Variable column length 2D Tensors?

Hello all,

I have an algorithm in MATLAB that uses “Cell-Datastructure” to store 1D column vectors of different lengths. Now, I want to code this in PyTorch. However, I did not find a way to use variable column length tensors for 2D matrices, as it seems like 2D PyTorch Tensors has to be strictly of “m x n” shape.

The only way that I can think of is using Python dictionaries, and holding various columns in each Dictionary entry.

My prime concern is of processing speed, as I would have lots of columns and python for loops are pretty slow when I apply them in this task. Is there a better way to do it without dictionaries and prefereably with PyTorch?

I don’t think there is any other valid approach besides manually looping over the dict/list or padding all tensors.
Nested tensors should solve this issue and might provide a clean way without padding. The progress can be tracked here.