OK, so when you are referring to predictions on the training set, you mean
predictions from a cross-validation on the training set. You can't get those
through the predict() method in RWeka.
You could try Vote or Stacking in WEKA to build your ensemble.
Is the _JAVA_OPTIONS environment variable not being read by Java when you
use Java from R?
From: wekalist-bounces(a)list.waikato.ac.nz <wekalist-
bounces(a)list.waikato.ac.nz> On Behalf Of Michael Hall
Sent: Saturday, 1 September 2018 9:59 AM
To: Weka machine learning workbench list. <wekalist(a)list.waikato.ac.nz>
Subject: Re: [Wekalist] RWeka training predictions
On Aug 31, 2018, at 4:42 PM, Eibe Frank
Have you tried ZeroR or IBk with a large value for K?
I tried different K values in training for IBk, didn’t really seem to
not sure what this will change or tell me though as far as getting actual
predictions for the training set?
I put some time in trying to figure out how to do this with Explorer. I
determine an ensemble there maybe and transfer it to RWeka.
But predictions appear to be output by fold if you do CV, I don’t see how
a split would work, using full training set just appeared to give me
classification again, no errors - I wish.
I’m thinking at the moment I’ll just try to write some simple java to do
classifications. I would prefer to get the full probability distributions
Then make the ensemble summing the probabilites
across classifiers and taking the max. This was what I wanted in RWeka,
worked against the test set for two classifiers although the ensemble did
worse than just RandomForest.
I would like to maybe add one more classifier to the ensemble and try
weighting the probabilities being summed. Being able to test with the
set there would be needed.
Off-topic but do you know of anyway to preallocate more memory for the
java jvm using RWeka? The data has a fairly large test set which has been
running into frequent memory related problems.
Wekalist mailing list
Send posts to: Wekalist(a)list.waikato.ac.nz
List info and subscription status: