search

fchollet / deep-learning-with-python-notebooks

Jupyter notebooks for the code samples of the book "Deep Learning with Python"
MIT License
17.95k stars 8.48k forks source link

Listing 7.20 Custom Callback (2nd edition) #222

Closed steval88 closed 1 year ago

steval88 commented 1 year ago

Dear Francois

there's a small typo in Listing 7.20 on page 190. In model.fit(), epochs is set to 10, but to be consistent with Figure 7.5, it should be epochs=1

Best

Stefano

steval88 commented 1 year ago

I was double-checking : it's actually not a typo, but if you run Listing 7.20 on jupyter without changing the epoch number, it yields the last training loss at epoch=9. However, you can retrieve the training loss from the plots which are created and stored in your local folder for each epoch. Steady state loss is around 0.07 at epoch=9 after 200-400 batches. Maybe adding some comments about the different behaviors (e.g. for small batch numbers for different epoch numbers) would have been interesting