Errata

The link for reading advanced transformer techniques online is not working, there is no preview available online on Google drive.

Note from the Author or Editor:
Thanks for your feedback. The online content is being finalized now, it will be up shortly at: https://homl.info/
(the old link https://homl.info/att will just redirect to this address).

Both locations mention PapersWithCode.com, which has now been discontinued and redirects to Hugging Face's trending papers section instead.

Note from the Author or Editor:
Great catch, thanks. I changed the link to point directly to HF papers.

In Figure 2-6, we can see housing.head() being called, whereas it should state housing_full.head(). This error is correct in the colab notebook, but present on the O'Reilly platform.

Note from the Author or Editor:
Good catch, thanks! I updated the figure (a screenshot of this cell in the colab notebook).

There is a language there saying: "We use the functools.partial() function (introduced in Chapter 11) to define DefaultConv2d..."

Well the functools.partial() was actually introduced in Chapter 10, therefore the correct version is: "We use the functools.partial() function (introduced in Chapter 10) to define DefaultConv2d..."

Note from the Author or Editor:
Good point, thank you, indeed the sentence should point to Chapter 10, not Chapter 11.

There is a reference there saying: "precision/recall curve may contain a few sections where precision actually goes up when recall increases, especially at low recall values (you can see this at the top right of Figure 3-6)"

In fact it should refer to the top left corner of Figure 3-6 so that the correct version is: "precision/recall curve may contain a few sections where precision actually goes up when recall increases, especially at low recall values (you can see this at the top left of Figure 3-6"

Note from the Author or Editor:
Good catch, thanks. Indeed, this sentence should point to the top left of Figure 3-6, not the top right.

The paragraph reads,

"... Since each prediction has a shape of [1, 1], we must use unsqueeze() again to add a batch dimension..."

Shouldn't it be,

"... to add a TIME dimension..." since the predictions' shape of [1, 1] is actually [batch, feature]?

In this specific example the result does not change because of the shape of the data but in its current form the sentence gives a false or confusing factual understanding of what's going on.

Note from the Author or Editor:
Thanks for your feedback, indeed we're adding a time dimension, not a batch dimension.

I think the line seen in the code,

>>> X = torch.cat([X, y_pred_one.unsqueeze(dim=0)], dim=1)

should be:

>>> X = torch.cat([X, y_pred_one.unsqueeze(dim=1)], dim=1)

since the predictions' (y_pred_one) shape of [1, 1] is already [batch, feature]. thus, unsqueezing should be done along time dimension, which is 1.

(In this specific example the result does not change because of the shape of the data but in its current form the sentence gives a confusing factual or conceptual understanding of what's going on.)

Note from the Author or Editor:
Great catch, thanks a lot. Indeed, it should have been `y_pred_one.unsqueeze(dim=1)` instead of `y_pred_one.unsqueeze(dim=0)`.
I fixed this in the book and the notebook.

the 2 links provided on page 19-20 to submit the errata are wrong on.

"We have a web page for this book, where we list errata, examples, and any additional information. You can access this page at homl_dot_info/oreilly-p"
&
"....if you find errors in the code examples (or just to ask questions), or to submit errata if you find errors in the text. "

Note from the Author or Editor:
Thanks for your feedback. The link is now fixed.

"Moreover, we make the discriminator untrainable by setting p.required_grad = False for each parameter p."

p.required_grad should be p.requires_grad.

Note from the Author or Editor:
Good catch, thanks! Indeed, it should be p.requires_grad, not p.required_grad. I fixed the book, so future reprints will be good.