model.get_topic_freq() show different count per topic than i get by using document_topics, _ = model.transform(data['text']) and filter the rows

Nobody has claimed this yet.

i have a 5 mil rows dataset of Reddit posts i am doing topic analysis on.
i fitted the model, all seems to make sense, the clusters i got make sense.

When I look at the cluster's sizes in model.get_topic_freq() it shows the counts for each topic ( how many documents are in the cluster)

however when i run document_topics, _ = model.transform(data['text']) and look at the rows in the dataset that correspond to a topic, it shows a different count...
for example for topic 4:
model.get_topic_freq() -> gives a count of about 3000
len(data[document_topics == 4])-> gives a count of around 6000

when i look at the text in those rows it is making sense ...again...the topic makes sense.... but the counts differ ...

why is this happening ... i can't understand it ??
sorry if this is a silly question, please if there is documentation that would help with this link to it ...i looked but can't seem to find anything that would clarify this issue...

thank you!

Start with the reported calls to model.get_topic_freq() and model.transform(data['text']), reproducing the differing counts on the Reddit dataset. Compare the topic assignments returned by each entry point and check the BERTopic documentation for the documented meaning of each result; done means the discrepancy is explained or confirmed as a defect.

Written by the indexing model from the issue text.