Getting errors 1) for a larger zeroshot_topic_list and 2) when tried to reduce outliers

[mjavedgohar]

Hi @MaartenGr ,

I need your help regarding the following issues.

1. I'm experiencing an error when using BERTopic (0.16.0) with zeroshot_topic_list on a small dataset. The error occurs when the number of topics provided in the list exceeds e.g. > 10. While the model works fine with fewer topics, Could you please assist me in resolving this issue?

Code:
`vectorizer_model = CountVectorizer(stop_words="english", ngram_range=(1, 4))

topic_model = BERTopic(language="english",
#top_n_words=15,
min_topic_size=10,
zeroshot_topic_list=zeroshot_topic_list,
zeroshot_min_similarity=0.6,
verbose=True,
representation_model=representation_model_chatgpt,
calculate_probabilities=True,
vectorizer_model=vectorizer_model
)

topics, probs = topic_model.fit_transform(docs)`

Error due to a large list of zeroshot_topic_list:
topics, probs = topic_model.fit_transform(docs)
File ".\lib\site-packages\bertopic_bertopic.py", line 448, in fit_transform
predictions = self._combine_zeroshot_topics(documents, assigned_documents, assigned_embeddings)
File ".\lib\site-packages\bertopic_bertopic.py", line 3540, in _combine_zeroshot_topics
merged_model = BERTopic.merge_models([zeroshot_model, self], min_similarity=1)
File ".\lib\site-packages\bertopic_bertopic.py", line 3153, in merge_models
new_tensors = tensors[new_topic - selected_topics["_outliers"]]
IndexError: index -2 is out of bounds for axis 0 with size 1

2. the following errors I am getting when tried to reduce outliers

a. When using;

new_topics = topic_model.reduce_outliers(docs, topics, probabilities=probs, threshold=0.05, strategy="probabilities")

Error:
new_topics = topic_model.reduce_outliers(docs, topics, probabilities=probs, threshold=0.05, strategy="probabilities")
File ".\lib\site-packages\bertopic_bertopic.py", line 2142, in reduce_outliers
raise ValueError("Make sure to pass in probabilities in order to use the probabilities strategy")
ValueError: Make sure to pass in probabilities in order to use the probabilities strategy

b. and When use;
new_topics = topic_model.reduce_outliers(docs, topics)

Error:
new_topics = topic_model.reduce_outliers(docs, topics)
File ".\lib\site-packages\bertopic_bertopic.py", line 2153, in reduce_outliers
topic_distr, _ = self.approximate_distribution(outlier_docs, min_similarity=threshold, **distributions_params)
File ".\lib\site-packages\bertopic_bertopic.py", line 1279, in approximate_distribution
topic_distributions = np.vstack(topic_distributions)
File "<array_function internals>", line 180, in vstack
File ".\lib\site-packages\numpy\core\shape_base.py", line 282, in vstack
return _nx.concatenate(arrs, 0)
File "<array_function internals>", line 180, in concatenate
ValueError: need at least one array to concatenate

Thanks

[MaartenGr]

It might have something to do with the version that you are using. Since then, there have been several fixes to zero-shot topic modeling. Fortunately, I just released 0.16.4. Could you try that version?

[mjavedgohar]

Thanks @MaartenGr,
BERTopic (0.16.4) resolved these issues.

But I am geeting the following error on topic_model.save when using embedding_model=openai_embedder.

Code:

vectorizer_model = CountVectorizer(stop_words="english", ngram_range=(1, 5))
openai_embedder = OpenAIBackend(client=client, embedding_model="text-embedding-ada-002", batch_size=100, delay_in_seconds=2)

topic_model = BERTopic(language="english",
                        embedding_model=openai_embedder,
			top_n_words=15,
			min_topic_size=10,
    			zeroshot_topic_list=zeroshot_topic_list,
    			zeroshot_min_similarity=0.6,
                        verbose=True, 
                        representation_model=representation_model_chatgpt,
			calculate_probabilities=True,
                        vectorizer_model=vectorizer_model
                        ) 


topics, probs = topic_model.fit_transform(docs)

df_topics=pd.DataFrame(topic_model.get_topic_info())

if not df_topics[df_topics["Topic"]==-1].empty: 
    new_topics = topic_model.reduce_outliers(docs, topics, probabilities=probs, threshold=0.05, strategy="probabilities")
      
    topic_model.update_topics(
        docs,
        topics=new_topics,
        vectorizer_model=vectorizer_model,
        representation_model=representation_model_chatgpt 
    )

    

topic_model.representation_model = None

topic_model.save("my_model")

Error:

topic_model.save(f'{destination_path}{filen}{middle_part_filename}{today}')
File ".\lib\site-packages\bertopic_bertopic.py", line 3311, in save
joblib.dump(self, file)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 555, in dump
NumpyPickler(filename, protocol=protocol).dump(value)
File ".\lib\pickle.py", line 487, in dump
self.save(obj)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 998, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 998, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 998, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 998, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 1003, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 603, in save
self.save_reduce(obj=obj, *rv)
File ".\lib\pickle.py", line 717, in save_reduce
save(state)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 560, in save
f(self, obj) # Call unbound method with explicit self
File ".\lib\pickle.py", line 972, in save_dict
self._batch_setitems(obj.items())
File ".\lib\pickle.py", line 998, in _batch_setitems
save(v)
File ".\lib\site-packages\joblib\numpy_pickle.py", line 355, in save
return Pickler.save(self, obj)
File ".\lib\pickle.py", line 578, in save
rv = reduce(self.proto)
TypeError: cannot pickle '_thread.RLock' object

[MaartenGr]

I'm not sure what exactly is creating this issue, pickle can be finicky. Have you tried safetensors instead?