Numpy's exp() is already vectorized, remove redundant vectorize() call.

[nico]

No intended behavior change. The vectorize() documentation points out
that vectorize() is not fast, and indeed after this change the time to
train one epoch is reduced from ~37.6s to ~10.5s on my laptop -- a 72%
reduction.

I haven't read more than Chapter 1 yet (in fact, I haven't completed Chapter 1),
so maybe something like this can be done in other places too.

[nico]

…looks like this is a dupe of #20.

[mnielsen]

Hi Nico,

Thanks for pointing this out.

It is, indeed, a dupe. I've been so busy with other things (including finishing chapter 6 of the book) I haven't had time to go back to this. With the final chapter done, I should get to this soon. It's more complex than it looks, since I also need to update some parts of chapter 1 and 3 of the book, and that requires considerable care. In the meantime I'll close this issue, & keep #20 open.