Topic retrieval using terms

[DobromirM]

Given that TopicData has access to topic_term_matrix which shows us the importance of each term for each topic, is there a method for retrieving the top N topics for a given term (or a list of terms)?

[x-tabdeveloping]

One way you could do that is by treating a word like a document and calling topic_data.print_topic_distribution("your word/phrase")
I didn't want to implement this necessarily for multiple documents at a time since it's not trivial how you would aggregate (you could average I suppose). If you need a utility function for this, I can implement something in the next release.

You can also manually get the indices of the highest ranking topics on some words like this:

import numpy as np

components = topic_data.topic_term_matrix
vocab = topic_data.vocab
topic_names = topic_data.topic_names
# or
components = model.components_
vocab = model.get_vocab()
topic_names = model.topic_names

your_words = ["dog", "cat", "horse"]
# We extract the word indices first
terms_to_ids = {word: i_word for i_word, word in enumerate(vocab)}
word_indices = [terms_to_ids[word] for word in your_words]

# Then calculate average topic importance for these words
topic_importance = np.mean(components[:, word_indices], axis=1)
top10_topic_indices = np.argsort(-topic_importance)[:10]

# Then print them
for i_topic in top10_topic_indices:
    print(topic_names[i_topic], ": ", topic_importance[i_topic])

[DobromirM]

I initally tried using topic_data.print_topic_distribution("..."), as you suggested, but I get the following error when I run it with only one or a small number of words:

Traceback (most recent call last):
  File "/newsgroups.py", line 15, in <module>
    model.print_topic_distribution("actually")
  File "/.venv/lib/python3.10/site-packages/turftopic/base.py", line 446, in print_topic_distribution
    columns, *rows = self._topic_distribution(text, topic_dist, top_k)
  File "/.venv/lib/python3.10/site-packages/turftopic/base.py", line 407, in _topic_distribution
    topic_dist = self.transform([text])
  File "/.venv/lib/python3.10/site-packages/sklearn/utils/_set_output.py", line 319, in wrapped
    data_to_wrap = f(self, X, *args, **kwargs)
  File "/.venv/lib/python3.10/site-packages/turftopic/models/keynmf.py", line 269, in transform
    return self.model.transform(keywords)
  File "/.venv/lib/python3.10/site-packages/turftopic/models/_keynmf.py", line 288, in transform
    W, _, _ = NMF(
  File "/.venv/lib/python3.10/site-packages/sklearn/decomposition/_nmf.py", line 1702, in _fit_transform
    W, H = self._check_w_h(X, W, H, update_H)
  File "/.venv/lib/python3.10/site-packages/sklearn/decomposition/_nmf.py", line 1228, in _check_w_h
    raise TypeError(
TypeError: H should have the same dtype as X. Got H.dtype = float32.

[x-tabdeveloping]

Hmm curious... My intuition is that this might be because the terms you are specifying do not occur in the keyword matrix, and thus NMF has not learned a representation for them.
Of course this should not result in an exception, and even if it does, it should be more descriptive than this, so I will try to get it to work a bit more seemlessly.

[DobromirM]

I thought the same but I get the same error even when I use a word from the model vocabulary.