Lesson1-Pets error_rate increases in last epoch in resnet50 learner

I am new to this MOOC and attempted to run lesson1 ipynb. I am aware that metrics will not be the same as in the video due to randomisation, but in the Resnet50 run’s 8th (last epoch) I am getting increased error_rate despite reduced losses (as shown in the screenshot below)

How should I proceed OR Do I need to make changes to the model (maybe reduce epochs)?

1 Like

Hi Medha, and welcome. I appreciate your being alert to details of the lesson. Reducing the number epochs to get a lower error rate is certainly a valid strategy. It’s called “early stopping” in the machine learning field.

The phenomenon where training loss goes down while error rate goes up is a consequence of overfitting. It (and remedies) will be explained in later lessons.

HTH, Malcolm

Hi Malcom, Thank you so much for the crisp response. I’ll move forward to the further lessons then. This helped a lot.