Overfitting in the Lesson 1

I was working through first lesson jupyter notebook, i noticed an error_rate increase during training
although train_loss is decreasing error_rate is increasing, is it Overfitting?

Generally I would say no. Your training loss is higher than your validation loss and both are still decreasing. I would run for a few more epochs and see if things still improve.

1 Like

Hi ahmadf hope all is well!

It is my understanding that if you watch the video again you may notice the specifics of learn.fit_one_cycle(4), you will also notice that at this point we are training the model with resnet34 (transfer learning). So training has not been completed yet. From my experience in training models it is not unusual for the error rate to fluctuate like this during training.

Cheers mrfabulous1 :smiley::smiley:

1 Like


I am getting fewer than 4 rows of results ( one or two rows). What could I be doing incorrectly?

Resolved - fit_one_cycle(4) wasn’t running because the cell was NOT marked as code.:smirk::slightly_smiling_face: