Topics over time: IndexError: index (68) out of range

Looks related to this.

The environment:

Windows 64-bit

bertopic 0.17.3 pypi_0 pypi
hdbscan 0.8.40 pypi_0 pypi
numpy 2.2.6 pypi_0 pypi
python 3.13.7 hdf00ec1_100_cp313 conda-forge
sentence-transformers 5.1.0 pypi_0 pypi
umap-learn 0.5.9.post2 pypi_0 pypi

The code:

    df = pd.read_csv(data_path)
    documents = df['text'].tolist()
    date_format = '%Y-%m-%d %H:%M:%S'
    df['date'] = pd.to_datetime(df['date'], format=date_format)
    df['date'] = df['date'].dt.date.astype(str)
    timestamps = df['date'].tolist()

    topic_model = BERTopic(verbose=True)
    topics, probabilities = topic_model.fit_transform(documents)

    topics_over_time = topic_model.topics_over_time(documents, topics, timestamps, nr_bins=20, global_tuning=True, evolution_tuning=True)

The error:

Traceback (most recent call last):
File "D:\Users\user\workspace\Customer-DS009_topic_modelling\src\run_bertopic_simple.py", line 64, in
topics_over_time = topic_model.topics_over_time(documents, topics, timestamps, nr_bins=20, global_tuning=True, evolution_tuning=True)
File "C:\Users\user.conda\envs\bertopic\Lib\site-packages\bertopic_bertopic.py", line 925, in topics_over_time
c_tf_idf = (global_c_tf_idf[selected_topics] + c_tf_idf) / 2.0
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "C:\Users\user.conda\envs\bertopic\Lib\site-packages\scipy\sparse_index.py", line 30, in getitem
index, new_shape = self._validate_indices(key)
~~~~~~~~~~~~~~~~~~~~~~^^^^^
File "C:\Users\user.conda\envs\bertopic\Lib\site-packages\scipy\sparse_index.py", line 288, in _validate_indices
idx = self._asindices(idx, N)
File "C:\Users\user.conda\envs\bertopic\Lib\site-packages\scipy\sparse_index.py", line 332, in _asindices
raise IndexError(f'index ({max_indx}) out of range')
IndexError: index (68) out of range

Reproduce the failure with the provided Windows environment and example code, starting at BERTopic's topics_over_time entry point and the _bertopic.py line 925 shown in the traceback. Trace how selected_topics is used with the global c-TF-IDF sparse matrix and compare it with the scipy sparse indexing error. Done means the example completes without the IndexError and produces the topics-over-time result.