01 - PyTorch Workflow Exercices - Different data set gives values completely off

[gaspardthrl]

Hi when doing the exercises I created a data set of 1000 values with step 1.

I use the same model we used in the course but get absurd losses

However when I create a data set of values from 0 to 10 with step 0.02 it converges fine.

I am using L1Loss and SGD with lr = 1e-3

Does someone have any idea of what is going on?

[mrdbourke]

Hi @gaspardthrl ,

Good question!

You're right, different models/datasets will get different values.

The models used in the course are often tuned (from my own experimentation) to get fairly good results on the given dataset.

So I understand correctly, are you using a different model/dataset?

Or are you saying your version of the same model/dataset are getting different results?

If you could share your code/where the differences are, that'd be a great help!