Weka Gaussian Process problems

Hi,
I am using the Weka Gaussian process and I am experienced the following problems.
1.- Does not matter the dataset always uses all memory available (even 8 Gb)
2.-The progress bar goes quickly up to 50% then freezes
3.-That causes Knime to freeze as well
4.-During next 1-2 hours memory usage slowly goes down to 600Mb. During all this time Knime uses between 10-20% of the CPU
5.-Once memory is low then gets calculated
Any idea of these strange behavior I really want to use this method
Thanks
Paco

Hi Paco,

I tried it myself and ran into the same problem! Urghh …

Then I tried it with Weka 3.7 itself (i.e. with Weka’s Explorer, see, e.g, here for further instructions) and the same thing happened: The UI froze until the the model training has been finished.

Next I took Weka 3.8 and tried it there: it still takes a long time to train the model but at least the UI didn’t freeze anymore. Thus, as soon as we update to Weka 3.8 the issue hopefully will be solved (though I can’t give any promises here when the update is going to happen).

As I work-around for now I hoped to be able to export the trained model from Weka 3.8 (which normally is possible), but for unknown reasons the ‘save model’ function is not available for Gaussian Process in the Weka Explorer.

Thus, the only suggestion I can make for now is that you could ask the Weka-guys how to export the Gaussian Processes-Model (if possible) from the Weka Explorer such that you can make use of it in KNIME as a workaround for the time being (weka models can be imported into KNIME with the “Weka Classifier Reader”-node).

Hope that helps! If you like, keep us updated in the thread here when you’ve got more insights.

Cheers,
Martin

2 Likes

Thanks Martin,
At the moment I think I can manage as my major focus is evaluating ML tools, like Knime, for my company
Thanks
Paco

Hi Paco!

If any questions, comments, suggestions after evaluation feel free to use this forum :wink:

Br,
Ivan