pytorchlightning version mismatched or missing some code

[priyanka36]

Why is this lr_scheduler_step() function not overridden in the script the pytorchlightning version requires this function to be overloaded as well

[quancs]

I used many versions of lightning, including the latest one, but I never override lr_scheduler_step. Does it report an error?

[priyanka36]

Yes i encountered in all versions of lightning i tried with every other version throws out the same error

[quancs]

The code is upgraded for lightning >= 2.0.0 now. And I just verified that the code is OK.

[quancs]

Yes i encountered in all versions of lightning i tried with every other version throws out the same error

You can post the error. See if I can help some.

[priyanka36]

So the error goes pytorch-lightning requires lr_scheduler to be overriden . By the way what is the default epoch number set for NBSS?

[quancs]

So the error goes pytorch-lightning requires lr_scheduler to be overriden .

You can post the error. Lightning doesn't require to override lr_scheduler_step, and implementing the configure_optimizers function is enough in lightning to configure an optimizer and lr_scheduler.

By the way what is the default epoch number set for NBSS?

100 for NBC2. For other archs, you can refer to our paper.

[priyanka36]

Sorry but I encountered that error . and that was at the end of epoch so I overrode it . Any way to change the epoch number ?

[quancs]

Any way to change the epoch number ?

You can use --trainer.max_epochs=100

[quancs]

Sorry but I encountered that error .

It's OK. There might be something that I didn't notice.

[priyanka36]

Thankyou. How long should we train the model. Does 10 epoch give as good of performance as 100?

[quancs]

Does 10 epoch give as good of performance as 100?

You can check the validation loss curve using tensorboard:

• Open tensorboard on log dir: tensorboard --logdir=logs --bind_all.
• Open tensorboard webpage in brower

In my experience, the epochs needed depends on task.

[priyanka36]

Thankyou