makcedward/nlp

:memo: This repository recorded my NLP journey.

nlp's Issues

ULMFiT from fastai?

Hi,

Just found your blog and I there are many useful information
I just found that you put ULMFiT under Openai in your readme table while according to I know, it is from fastai.
Thanks for compiling and sharing your knowledge.

returns lda instead of lsa

nlp/sample/nlp-lsa_lda.ipynb

Line 105 in 2f12277

"tfidf_vec, svd, x_train_lda, x_test_lda = build_lsa(x_train, x_test)"

In the above line, you assign the result of the LSA to the variables x_train_lda, x_test_lda. Shouldn't it be lsa in both cases?

InferSent error (help needed)

Hi, I am getting an error while generating InferSent embeddings. The error is as follows, with details at the end of this email

UnicodeDecodeError: 'utf-8' codec can't decode byte 0x86 in position 11: invalid start byte

The error occurs after I run infer_sent_embs.build_vocab(x_train, tokenize=True) .

Note that I ran your code in Google Colab. Also note that the links to InferSent in the python file infersent.py also need to be updated (expired links).

The new links are

INFERSENT_GLOVE_MODEL_URL = 'https://dl.fbaipublicfiles.com/infersent/infersent1.pkl'
INFERSENT_FASTTEXT_MODEL_URL = 'https://dl.fbaipublicfiles.com/infersent/infersent2.pkl'

`

UnicodeDecodeError Traceback (most recent call last)
in ()
----> 1 infer_sent_embs.build_vocab(x_train, tokenize=True)
2 x_train_t = infer_sent_embs.encode(x_train, tokenize=True)
3 x_test_t = infer_sent_embs.encode(x_test, tokenize=True)

3 frames
/usr/lib/python3.6/codecs.py in decode(self, input, final)
319 # decode input (taking the buffer into account)
320 data = self.buffer + input
--> 321 (result, consumed) = self._buffer_decode(data, self.errors, final)
322 # keep undecoded input until the next call
323 self.buffer = data[consumed:]

UnicodeDecodeError: 'utf-8' codec can't decode byte 0x86 in position 11: invalid start byte
`

Doc2Vec

@makcedward I am trying to retrieve similar documents from the given document. Here is the code snippet:

x_train_t = doc2vec_embs.encode(documents=x_train)
x_test_t = doc2vec_embs.encode(documents=x_test)

def similiar_docs(doc2vec_embs, test_sample):
sims = doc2vec_embs.model.docvecs.most_similar([test_sample], topn=1)
for s in sims:
print(x_train[s[0]])

test_sample = x_test_t[0]
print(x_test[0])
similiar_docs(doc2vec_embs, test_sample)

However, the retrieved docs aren't similar. Am I missing something here?

No module named aion. Issue while importing aion

Hello,

Please help me understand how to execute "from aion.util.spell_check import SymSpell". I tried to use the sys,os but got lost.

About DataSet

How can i get glove.6B.50d.vec which is imported in sample/nlp-word_mover_distance.ipynb of this repository

Where did you pull the code from?

I can't find and use the exact code that you have used. any idea on that?

nlp-character_embedding.ipynb

Hi, thank you for the awesome code on character embedding model, I had a lot of fun playing with it.
One little suggestion on the code, in CharCNN, def build_char_dictionary, you put chars = list(set(chars)). This will mess up the chars order in char_dict. Every time I start a new notebook, the chars' order will be different, therefore result in a different dictionary. What happened to me is I tried to load my trained keras model in a new notebook and found out that the model is not working. In the end I figured out it is because of my char_indices in preprocess step in totally different than old one. I didn't save the old char_indices before so I have no choice but to retrain the model, lol.

Cell #31 ValueError master/sample/nlp-model_interpretation_shap.ipynb

explainer = shap.DeepExplainer(pipeline.model, encoded_x_train[:10])
shap_values = explainer.shap_values(encoded_x_test[:1])

x_test_words = prepare_explanation_words(pipeline, encoded_x_test)
y_pred = pipeline.predict(x_test[:1])
print('Actual Category: %s, Predict Category: %s' % (y_test[0], y_pred[0]))

shap.force_plot(explainer.expected_value[0], shap_values[0][0], x_test_words[0])

RETURNS:

ValueError: Dimensions must be equal, but are 10 and 100 for '{{node gradient_tape/functional_1/global_max_pooling1d/truediv_1}} = RealDiv[T=DT_FLOAT](gradient_tape/functional_1/global_max_pooling1d/sub_1, gradient_tape/functional_1/global_max_pooling1d/sub)' with input shapes: [10,512], [10,100,512].

No module named aion. Issue while importing aion

Hello Edward,

Thank you for the great article and detailed blog post about ELMO. I used one of the resources you posted for ELMO in keras before, I thought that implementation lacked some things, like ability to normalize vectors, your implementation is certainly better than the one in resources.

When I tried to import aion, I'm facing an error which says "ModuleNotFoundError: No module named 'aion'". I know that this is a fairly common packaging issue, but I wasn't able to resolve it, I didn't find any reliable online source either to be able to fix it. Please let me know if you have faced such an issue at all, any pointers will be appreciated.

I was able to successfully download aion by "pip install aion"

I'm using:

Python 3.6.7

Windows 10

Thanks

Cosine similarity typographic error

The Cosine Similarity code has a typographic error.

States transfaormed_results[i] and should be transformed_results[i]