reset optimizer for multiple runs

[aRI0U]

Hi,
I need to run multiple similar trainings in order to evaluate the average performance of my model, so I need to reset the weights of my network as well as the state of my optimizer, scheduler, etc.
I use the same LightningModule for all runs but a new Trainer for each run. I manually reset the weights of the network between each run so there is no issue here, however I don't know if the optimizers are reset when I pass the LightningModule to a new Trainer. If not, does anyone know how to reset them? It seems that I can access the optimizer with self.optimizers().optimizer but I don't know to what extent I'm able to overwrite it easily.
Thanks a lot!

[carmocca]

I don't know if the optimizers are reset when I pass the LightningModule to a new Trainer

It's not that we reset them manually, configure_optimizers is called again in the next fit call.

but I don't know to what extent I'm able to overwrite it easily.

You should be able to modify their state

[aRI0U]

Ok so basically each time I call fit new optimizers are created right?

[carmocca]

yes