Chapter 4 MNIST_basics : Extremely high training and validation loss with no change in accuracy

darshil_fcbayern · August 28, 2021, 4:15am

While replicating the books code, I got such results while fitting a learner, what went wrong here?

kBodolai · August 29, 2021, 4:27pm

Hi,

could you share a copy of the colab notebook reproducing this behaviour?

K.

darshil_fcbayern · August 30, 2021, 9:48am

Hi K, thanks for responding,
I found the issue after going through the code few times, I included sigmoid in my linear1 function and hence did not keep it in MNIST_loss function as I did not know learner requires sigmoid to be in the loss function.