Saving a dbscan model

[jricgar]

Hi all,

I want to train a dbscan clustering model (with dbscan node or optics node) in order to use it later. Until now I supposed I have to save the results of my training process as a distance model because neither of these nodes gives you a clustering model, but only a distances one.

Find attached a screenshot of my test workflow. I’m not sure about the optics configuration. DBSCAN node is working, but results are not ok… regular data is tagged as noise.

I get epsilon value from model table as variable (it’s a column inside it), this is the reason you can see all of these variable lines.

Maybe someone here could provide me an example of using a previously saved distance model in order to cluster new samples. Or highlight me any error…

Thank in advance.

[Iris]

Hi
the problem is that you can not apply a density based model.
The disadvantage of them is that they don’t have representatives like a k-means does. So a density based clustering is calculated on all data points to split them in parts. New data points might change the model.

The only way to save a model is to save the assignment of the data points together with the cluster. This is the definition of the clustering.

Afterwards new data points could be assigned to the cluster of the closest data point.

This is why you cannot save this model.
Best, Iris