I am trying to run PCA on a dataset of 40 examples with 971 attributes with
real number values. I keep getting an outOfMemoryError even when I increase
the values of -mx and -oss to 100000000 etc. I do this modification in the
CLI, then I open the GUI interface and run the experiments, and still get the
error and no results.
Currently I can only get PCA to work on a reduced dataset of half the examples
(485) I assume that Weka's implemention of PCA can only handle a upper limit
of 512 attributes but if I'm wrong in that regard let me know.
Also, previous versions of Weka allowed an AttributeselectionFilter in the
preprocessing stage allowing for PCA. Then the resulting set could be saved as
a different .arff file. newer versions only allow attribute selection in the
classification stage under the meta folder but there seems to be no way to
save the resulting principal components set.
I Don't mind using an older version but if there is a way to do it in 3.4.1
that'd be great.
Also if there is a way to run PCA on all my attributes it would be even
Any help would be greatly appreciated,
Show replies by thread