On the web page (source here) the license is stated as CC-BY-SA 3.0 (Creative Commons, By Attribution, Share Alike 3.0) yet in the paper (here) the license is CC-BY-NC-SA 3.0 (Creative Commons, By Attribution, Non-Commercial, Share Alike 3.0).

If the intent is to restrict commercial re-use, then the text on the web page should be changed to CC-BY-NC-SA 3.0 to reflect what license the PDF document is under. If the intent is to allow for commercial re-use and to make the document libre/free in the open source sense, the license should be changed to CC-BY-SA 3.0 in the PDF document.

Hi,

Thanks for the paper, it's a really nice lecture.

This link is broken (http 404)
I've searching that implementation in the viberary repo and I think it might be this notebook?

https://github.com/veekaybee/viberary/blob/main/src/notebooks/04_word2vec.ipynb

In cell 17, this variable is used without having been previously defined.

Hello, why does it say that we optimize theta and Pi on page 43 eq(11). I only see theta as a parameter

Loving your document so far, Vicky! (filing this as I read)

In the tf-idf example on page 33 the objects dict_a and dict_b appear all of a sudden. I could not derive from the example what they represent and had to open the total on example on github to understand. Even though it is a truncated implementation I think some explanation on these objects would enhance the example. If nothing else, by just adding a code comment what they are / where they are coming from.

Hi! I'm working through the embeddings write up and it's a fantastic read.

In my copy that I downloaded a little while ago, the hyperlink for "Gensim" on page 44 just goes to the first page of the document. I went to make a pr, but this line looks to have the correct url. I downloaded a fresh copy from the main page, but the link still goes to the top of the document and doesn't have hyperlink format.

I think the main site needs to get a fresh version of the pdf?

Hello!

Thanks for writing the book - the stuff I've read so far (the intro) seems very well explained and easy to read!

However, I'm trying to read it on Kindle Paperwhite and everything is just tiny, as the PDF pages are flowed for A4 pages.
Since it appears to be written in TeX, could you please generate a Kindle-friendly version? The screen dimensions are closest to B7 (and please use narrow margins, the device has a large bezel itself :) ).

Thanks again in any case!

I believe formula 7 should rather be a single vector: [12, 2, 5].
We start to have tensors when stacking embeddings of different users. Am i missing something?

It looks like there's a cut-and-paste mishap in figure 26, starting at line 1202 of the tex:

tfidf_df = pd.DataFrame(vector.toarray(), index=text_titles, columns=vectorizer.get_feature_names_out())

tfidf_df.loc['00_Document Frequency'] = (tfidf_df > 0).sum()
tfidf_df.T

# How common or unique a word is in a given document wrt to the vocabulary
dreams_langstonhughes    quote_william_blake    00_Document Frequency
bird                    0.172503              0.197242              2.0
broken                  0.242447              0.000000              1.0
cannot                  0.242447              0.000000              1.0
die                     0.242447              0.000000              1.0
\end{minted}

I believe there may be a bug in the idf function in https://github.com/veekaybee/what_are_embeddings/blob/main/notebooks/fig_24_tf_idf_from_scratch.ipynb

Namely, the number of documents in which a term appears does not seem to be actually computed, just initialized to 0.

You could see that the results are wrong by looking at the printed tf-idf values for document 0 for words "a" and "bird": the printed values are equal, which doesn't make sense: although tf for both of them is the same (since they each appear in the document0 once), idf should be different, since "a" appears only in document0, and "bird" appears in document0 and document1.

I really appreciate your paper on embeddings, but I feel that a paragraph in the introduction that links back to the book's website along with a version indicator would help me establish if the PDF I have is current and can be read as is, or if it is out dated and should be deleted and re-downloaded.

Also, it helps if I send the PDF to someone instead of just sending them a link to your website. =)