I would like to compute a decomposition of a tensor with (a small number of) missing entries. In the contrib.decomposition.tensor_train_cross function(s) is it possible to reject the sampling of fibers that contain masked entries?

solution

I would like to add a simple 'mask' keyword to the function, which is a tensor with the masked entries.
When selecting the fibers/columns that will be sampled, reject any that contain masked entries.

alternatives

I would use the tensorly.contrib.sparse backend, however, it is not functional at the moment (see issue #499 ).