Predicting the label without inserting the datapoint into model

[shivankar-p]

I was trying to predict the output label without inserting into model(I have already trained the model on my data). It always gives Nan. Is there any specific reason for this kind of behaviour?

I thought it should give the closest existing cluster if not fitted with the data point.

[DataOmbudsman]

It's a matter of definition I think. I see what you expect but I can also argue that clustering is about arranging already existing data points into groups. Note that in scikit-learn's DBSCAN there's no predict method, only fit and fit_predict.

As a workaround you can maybe insert the new point, get its cluster label, then delete it.