search

arsedler9 / lfads-torch

A PyTorch implementation of Latent Factor Analysis via Dynamical Systems (LFADS) and AutoLFADS.
https://arxiv.org/abs/2309.01230
Other
85 stars 20 forks source link

lfads-output did not include train_inds and valid_inds. #11

Closed Riverside-ms closed 10 months ago

Riverside-ms commented 10 months ago

Hi Andrew,

I was able to proceed through the tutorial and got the lfads-output. However, the lfads-output files did not contain train_inds and valid_inds as variables, and I got an error in "Load LFADS factors and rates from output files".

I remembered that I had defined valid_inds as a multiple of 5 in "Saving the data for lfads-torch". So instead of reading train_inds and valid_inds from lfads-output, I defined them as not multiples of 5 and multiples of 5, thus avoiding the error. Is this the correct way to do this?

The lfads-output downloaded at the beginning of 3_evaluation.ipynb contained train_inds and valid_inds, but not followed the multiple of 5 rule.

Thanks.

arsedler9 commented 10 months ago

Thanks for pointing this out! You're right, the data indices were not saved by the 1_data_prep.ipynb notebook. I fixed this in the above PR. Your approach of manually recalculating the indices should also work. As you mentioned, the provided LFADS output files used a different train / valid split.