map_probabilities

[bexy-b]

Hi there,

I had a question about the map_probabilities function. It seems to only update the probabilities attribute if len(probabilities.shape) == 2 but my question is when is this condition actually met?

As far as I'm aware the shape should always be (n_docs, ) if running a "regular" topic model since the probabilities are given for the assigned topic per document only, or if you set calculate_probabilities=True then it'll be (n_docs, n_topics) right?

I want to understand if/when the probabilities per document are updated, but as far as I understand it this would only be recalculated if you called the hdbscan model again. The only thing that changes is that you're saying document X with probability p is now mapped to topic Y instead of topic Z but doesn't the probability p stay the same?

[MaartenGr]

When you run .shape on an array, it will give you back a tuple that contains one or more values. So when you do len(probabilities.shape) it will actually tell you how many dimensions that array has instead of the exact dimensions. Therefore, when you have two dimensions (regardless of their size), len(probabilities) will be 2.