function [trconf tsconf cl] = train_weka_classif_affective( train, test, classf, varargin )
%TRAIN_WEKA_CLASSF Train and test a Weka classifier on a data set
% [TRCONF TSCONF CL] = TRAIN_WEKA_CLASSF( TRAIN, TEST, CLASSF, ARG1, ARG2... )
% [TRCONF CL] = TRAIN_WEKA_CLASSF( TRAIN, [], CLASSF, ARG1, ARG2... )
% CL = TRAIN_WEKA_CLASSF( TRAIN, [], CLASSF, ARG1, ARG2... )
%   TRAIN      '.arff' file name for the training set, or a weka.core.Instances
%                object
%   TEST       '.arff' file name for the test set
%   CLASSF     Full name of the classifier Java class ('weka.classifiers...')
%   ARG1...    Additional parameters to the classifier; strings.
%
%   TRCONF     Confusion matrix for the training set. Columns are
%                classification, rows are the true class.
%   TSCONF     Confusion matrix for the test set.
%   CL         Trained classifier

    if nargin < 3
        error( 'Not enough arguments' )
    end
    dotest = ~isempty( test );
    doeval = dotest || (~dotest && nargout > 1);
    if (~dotest && nargout > 2) || (dotest && nargout > 3)
        error( 'Too many output arguments' )
    end

    cl = javaObject( classf );
    if length(varargin) > 0
        cl.setOptions( varargin{1} )
    end
    if ischar( train )
        tdata = weka.core.Instances( java.io.BufferedReader( java.io.FileReader( train ) ) );
        tdata.setClassIndex( tdata.numAttributes() - 1 ); % Class is the last attribute
    elseif isjava( train ) && isa( train, 'weka.core.Instances' )
        tdata = train;
    else
        error( 'Invalid type of first argument' )
    end
    cl.buildClassifier( tdata )
    if doeval
        evtr = weka.classifiers.Evaluation( weka.core.Instances( tdata, 0 ), [] );
        evtr.setPriors( tdata )
        if dotest
            evts = weka.classifiers.Evaluation( weka.core.Instances( tdata, 0 ), [] );
            evts.setPriors( tdata )
        end
        evtr.evaluateModel( cl, tdata );
        trconf = evtr.confusionMatrix();
        clear evtr
        if dotest
            tdata = weka.core.Instances( java.io.BufferedReader( java.io.FileReader( test ) ) );
            tdata.setClassIndex( tdata.numAttributes() - 1 ); %% Class is the last attribute
            evts.evaluateModel( cl, tdata );
            tsconf = evts.confusionMatrix();
        else
            tsconf = cl;
        end
    else
        trconf = cl;
    end

end % train_weka_classif_affective()