Confusion on page 423

[baifanhorst]

At the bottom of this page, it says that the model has 100% accuracy on the training set, and 95% accuracy on the validation set. However, the left panel of Fig13.4 shows that the model's accuracy on the training set is about 90% and 85% on the validation set. So there is a discrepency between the figure and the description in the text.

[baifanhorst]

I tried several times. The model in the book could give the described accuracy, but not always. There is some stochasticity in the result. Not a big issue, though.

[rasbt]

Good observation, I just double-checked the TensorFlow edition from 2019, where I got better results. I must have forgotten to update the text here. I will add this as a note to the errata. The interesting question is also why the PyTorch accuracies are lower. Probably not optimal hparam settings.