dpeerlab / spectra

Supervised Pathway DEConvolution of InTerpretable Gene ProgRAms
MIT License
134 stars 17 forks source link

Monitoring training #18

Open cpuritz opened 1 year ago

cpuritz commented 1 year ago

Hi,

Thank you for creating this really interesting project. I'm eager to use it in my work. One thing that would be really useful is if there was some way to easily examine training history to actually confirm convergence. Also, if some kind of early stopping could be implemented, that would be very helpful as Spectra takes a while to run.

Thank you, Connor

russellkune commented 1 year ago

Hi Connor,

Thanks! Have you tried setting verbose to True in the est_spectra or training function? You can monitor training this way. You can also alter the lr_schedule parameters (input to est_spectra as a kwarg). It takes in a list, e.g. est_spectra( ... lr_schedule = [1e-2,1e-3, ...], verbose = True).

I'll alter the behavior under verbose = True to also print losses, so I'll keep this open for now