fit_transform tries to access embedding_model if representation_model is not None

[rasantangelo]

Have you searched existing issues? 🔎

• I have searched and found no existing issues

Desribe the bug

I was using BERTopic on a cluster of queries with my own embeddings (computed on a model that is hard to pass as a parameter) and it was working as expected.
After trying to use representation_model = KeyBERTInspired() and adding representation_model=representation_model to BERTopic as a parameter. I got this error :

AttributeError                            Traceback (most recent call last)
      [1] representation_model = KeyBERTInspired()
      [2] topic_model = BERTopic(
      [3]     calculate_probabilities=True,
      [4]     min_topic_size=1
      [5]     embedding_model=None,
      [6]     representation_model=representation_model,
      [7] )
----> [8] topics, probs = topic_model.fit_transform(corpus, np.array(corpus_embeddings))
      [9] topic_model.get_topic_info()

File ~/query_analysis/bertopic_env/lib/python3.11/site-packages/bertopic/_bertopic.py:493, in BERTopic.fit_transform(self, documents, embeddings, images, y)
    [490]     self._save_representative_docs(custom_documents)
    [491] else:
    [492]     # Extract topics by calculating c-TF-IDF
--> [493]     self._extract_topics(documents, embeddings=embeddings, verbose=self.verbose)
    [495]     # Reduce topics
    [496]     if self.nr_topics:

File ~/query_analysis/bertopic_env/lib/python3.11/site-packages/bertopic/_bertopic.py:3991, in BERTopic._extract_topics(self, documents, embeddings, mappings, verbose)
   [3989] documents_per_topic = documents.groupby(["Topic"], as_index=False).agg({"Document": " ".join})
   [3990] self.c_tf_idf_, words = self._c_tf_idf(documents_per_topic)
-> [3991] self.topic_representations_ = self._extract_words_per_topic(words, documents)
...
   [3680]        "Make sure to use an embedding model that can either embed documents"
   [3681]        "or images depending on which you want to embed."
   [3682]    

AttributeError: 'NoneType' object has no attribute 'embed_documents'

Reproduction

from query import Query
import json 
import numpy as np
from sklearn.cluster import DBSCAN
from bertopic import BERTopic
from openai import OpenAI 
from bertopic.representation import KeyBERTInspired

representation_model = KeyBERTInspired()
topic_model = BERTopic(
    calculate_probabilities=True,
    min_topic_size=15,
    embedding_model=None,
    representation_model=representation_model,
)
topics, probs = topic_model.fit_transform(corpus, np.array(corpus_embeddings))
topic_model.get_topic_info()

BERTopic Version

0.16.4

[MaartenGr]

You will need to set embedding_model to an embedding model since KeyBERTInspired actually needs to create word embedings.