much slower s/it when run from CLI

[r-shruthi11]

Thanks for creating and sharing this codebase. I noticed a big difference in how both data loading and fitting it/s or s/it when I call load_data or fit_model from the command line as opposed to within a jupyter notebook. Could this be related to tdqm printing progress to stdout vs. in the notebook?
(Also please let me know if there is a different forum to raise issues or discuss bug fixes) - SR

[calebweinreb]

Can you clarify what, exactly, is slower? The actual execution of the code? Or the rate of progress updates by tqdm? If it's possible to provide a minimal example to reproduce the issue and an example of the output, that would be helpful too.

[r-shruthi11]

Hmm, good question. tqdm reports a value iterations/s or s/iteration. For example, the when you call fit_model there is a progress bar that reports this. This is what appears to be slower when I run this or other similar functions from the CL.
After some profiling, I realized that it was the plot_progress function that was slowing things down. Turning that off speeds things up.

[calebweinreb]

Got it! Did you also find that load_data was slower, though? That would still be unexplained

[r-shruthi11]

Ack, you're right. load_data was also slower.