From: Steven Shack <firstname.lastname@example.org<mailto:email@example.com>>
Reply-To: "Weka machine learning workbench list."
Date: Saturday, 27 April 2013 5:39 AM
To: "Weka machine learning workbench list."
Subject: [Wekalist] Reloading learnt model.
Hello Weka list,
I have a bayesnet model which took about three days to learn. I'd like to be able to
reload this model
and test it against other datasets, as well as reuse it in the knowledge flow. Is this
I can see in the explorer you can load the model, but each time I go to try and test with
the learnt model it tries
to relearn the model. Obviously a three day learning/test process is longer than I'd
So what can I do here?
If you've saved the model out to a file then it can be reloaded into the Classify
panel of the Explorer by right clicking over the history list and choosing "Load
model". If a separate test set has been configured as the evaluation option then you
should see the "Re-evaluate model on current test set" menu option become
In recent versions of the KnowledgeFlow in the development version of Weka a model can be
loaded from a file and used to score test sets. You choose a the classifier component that
matches the type of model that you wish to load and then connect just a
"testSet" connection to it (from a TestSetMaker component). In this case when
you open the Classifier components dialog you will see a "More options" section
at the bottom that contains a "Load model from file" field.
Another question is to ask if there's any work going into parallelizing or speeding up
the bayesnet classifiers (The search algorithms in particular).
A quick profile of the running code showed the majority of the time spend in
calcNodeScorePlain in the local
search code. That code looked to be doing a lot of boxing and unboxing of data, with some
indirection. Can this be GPU
I'll let Remco answer this if he's reading the list.