Bertopic with embedding: unable to use find_topic

Question

I've used BERTopic with success for the following tasks: get topics, visualise (topics, barcharts, documents ...) and DTM (extended to get area plot with considerable success).

However, I am unable to use the find_topics() function

(There are a few others I'm struggling with, which I'll post as new questions so as not to conflate this one).

I get an error message indicating that I'm using embedding (which is true).

# Prepare embeddings using default 'sentence embedding'
sentence_model = SentenceTransformer("all-MiniLM-L6-v2")
embeddings = sentence_model.encode(docs_bert, show_progress_bar=True)

Trying to solve that, I have tried to instantiate a new model without embedding

model_ngram_embed2 = BERTopic(embedding_model=embeddings)

but it then throws an error:

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

I need to instantiate before I can fit_transform the model to my doc (text corpus), after which I would then be able to find_topics().
How do I go about that? What should be done?

Regarding find_topics(), I've read allowing precomputed embeddings in bertopic.find_topics() issue
NB: Python 3.8.8 | IPython 7.31.1 | BERTopic 0.11.0

Answer 1

To resolve this, I did the following.
Firstly, I updated BERTopic to 0.12.0. This did not resolve on its own.
What works is excluding seed_topics: i.e. not using semi-supervised mode.

%%time
## instantiate BERTopic with bigram
model_ngram_noembed2 = BERTopic(top_n_words=10, min_topic_size=5, n_gram_range=(1,2), nr_topics=41, verbose=True)  
                     #, embedding_model=embeddings, seed_topic_list=seed_topic_list)

## Fit the models on doc with embedding, generate topics, and return the docs with topics
topics_bert_noembed2, probs_noembed2 = model_ngram_noembed2.fit_transform(docs_bert_dtm)

For completeness, the rest are

## Explicate 'define' within the topic model
keyterm_explore = 'define'

#pd_keyterm_explore = pd.DataFrame(topics_bert.find_topics(keyterm_explore,))
'''
Exception: This method can only be used if you did not use custom embeddings.'''

pd_keyterm_explore = pd.DataFrame(model_ngram_noembed2.find_topics(keyterm_explore), 
             columns=model_ngram_noembed2.find_topics(keyterm_explore)[0],
            dtype='float').round(decimals=2)

print(f'Exploring {keyterm_explore}: \n{pd_keyterm_explore}')

    [out]
    Exploring define: 
        21     12     19    7     3 
    0  21.00  12.00  19.00  7.00  3.00
    1   0.37   0.36   0.34  0.32  0.32