Hello Abhishek,

I see that in the engine, scheduler is being called every (b_idx + 1) / self.acc_steps, that is effectively every batch if acc_steps = 1

But in various implementations, I have seen that sechuler.step() is called after the validation step. Can you provide some insight as to why you have used a scheduler in this manner?

Another question, let's say I am using ReduceLROnPlateau which requires I pass a metric in step.step(). How do we use engine related to that?

Thanks in advance 😄

I have created a function to log the gradients if they are vanishing or exploding. Would like to contribute the function here.

@abhishekkrthakur Let me know if I can create a PR for the same that is if you want to add this feature.

I followed your tutorial Detecting Skin Cancer (Melanoma) With Deep Learning and i cant make the code to work. I keep getting this type of error .... TypeError: object of type 'ClassificationDataLoader' has no len().... I did download this github repository and did put it in python site-packages to import it. Maybe thats what i did wrong?

Hi Abishek,
The image dataloader is not compatible with GPU, there is a cannot unpack non-iterable object issue
Thanks for your help