practical-nlp/Ch4/03_Word2Vec_Example.ipynb

[sai-teja-ponugoti]

practical-nlp/Ch4/03_Word2Vec_Example.ipynb

[varunp2k]

Thank you @sai-teja-ponugoti for bringing this to our notice.
The fix to this most likely would be just having
np.mean(w2v_model[token],axis=0)
instead of the current
w2v_model[token]

Would you like to send a PR to fix this issue?

[brianalexander]

Hi @varunp2k,

I believe the fix would be a modification of

feats.append(feat_for_this)
to
feats.append(feat_for_this/count_for_this if count_for_this > 0 else feat_for_this)

In the example, we're categorizing the tweet by taking the average of all of the word vectors for each word in the sentence. This also makes sense because previously count_for_this is included in the code, but never ends up being used.

count_for_this if count_for_this > 0 else feat_for_this is needed to handle cases where there are no token matches and avoid dividing by zero.

Here is the updated cell in its entirety:

def embedding_feats(list_of_lists, dims=300):
    zero_vector = np.zeros(dims)
    feats = []
    for tokens in list_of_lists:
        feat_for_this = np.zeros(dims)
        count_for_this = 0
        for token in tokens:
            if token in w2v_model:
                feat_for_this += w2v_model[token]
                count_for_this += 1
        feats.append(feat_for_this/count_for_this if count_for_this > 0 else feat_for_this)
    return feats

train_vectors = embedding_feats(texts_processed)
print(len(train_vectors))

[varunp2k]

Looks good @brianalexander
Please send a PR

[sai-teja-ponugoti]

Thank you @sai-teja-ponugoti for bringing this to our notice.
The fix to this most likely would be just having
np.mean(w2v_model[token],axis=0)
instead of the current
w2v_model[token]

Would you like to send a PR to fix this issue?

@brainalexander have you sent one already?

[varunp2k]

@sai-teja-ponugoti
A PR stands for Pull Request. This article will explain what it is in a detail.
https://docs.github.com/en/github/collaborating-with-issues-and-pull-requests/about-pull-requests
Brainalexander is yet to send a PR.
You can check the active PR's for this repo under the pull requests section right next to the issues. The link to that is here.