WEKA Forecasting Plugin - weka

Question

My question is related to Weka Forecasting Plugin. I'm trying to predict the value of sin(x) in a prediction horizon of 6 using Weka Forecaster, the algorithm is SMOReg, and I'm trying to use different lags ( 1 - 24 ). I'm using a loop for that. The program I wrote is working perfectly but the problem is the results I obtain from my program are different than using Weka GUI directly and I don't know why? Any ideas?
I made sure to use the same "options" of the algorithm but still results are different.
Here's the code:
import java.io.*;
import java.util.*;
import java.util.List;
import weka.core.Instances;
import weka.classifiers.Evaluation;
import weka.classifiers.functions.*;
import weka.classifiers.rules.*;
import weka.classifiers.meta.*;
import weka.classifiers.trees.*;
import weka.classifiers.evaluation.NumericPrediction;
import weka.classifiers.timeseries.WekaForecaster;
import weka.classifiers.timeseries.core.TSLagMaker;
import weka.classifiers.timeseries.core.TSLagMaker.Periodicity;
import weka.classifiers.timeseries.eval.TSEvaluation;
import weka.classifiers.timeseries.eval.ErrorModule;
import weka.core.converters.ConverterUtils.DataSource;
public class TimeSeriesSin {
static Scanner console = new Scanner(System.in);
public static void main(String[] args) throws FileNotFoundException{
try {
String pathToSinData = "C:/Users/khouloud/Desktop/AUSmaster/thesis/summer/chapter1/sin_10.arff";
PrintWriter out = new PrintWriter("C:/Users/khouloud/Desktop/AUSmaster/thesis/summer/chapter1/try_sin_10.txt");
DataSource source = new DataSource(pathToSinData);
Instances diab = source.getDataSet();
for(int i = 1; i <= 24; i++) {
WekaForecaster forecaster = new WekaForecaster();
forecaster.setFieldsToForecast("sin");
SMOreg tr = new SMOreg();
String options = "weka.classifiers.functions.SMOreg -C 1.0 -N 0 -I \"weka.classifiers.functions.supportVector.RegSMOImproved -L 0.001 -W 1 -P 1.0E-12 -T 0.001 -V\" -K \"weka.classifiers.functions.supportVector.PolyKernel -C 250007 -E 1.0\"";
tr.setOptions(weka.core.Utils.splitOptions(options));
forecaster.setBaseForecaster(tr);
forecaster.getTSLagMaker().setMinLag(1);
forecaster.getTSLagMaker().setMaxLag(i);
forecaster.buildForecaster(diab, System.out);
forecaster.primeForecaster(diab);
List<List<NumericPrediction>> forecast = forecaster.forecast(6, System.out);
TSEvaluation eval = new TSEvaluation(diab, 0.3);
eval.setHorizon(6);
eval.setEvaluateOnTestData(true);
eval.setEvaluateOnTrainingData(false);
eval.evaluateForecaster(forecaster, System.out);
out.println("lag= " + i);
out.println(eval.toSummaryString());
}
out.close();
} catch (Exception ex) {
ex.printStackTrace();
}
}
}

Answer 1

I got the same problem for forecasting three different variables. I used MLPRegressor and SMOreg.
I checked the package version of the pdm-timseriesforecasting package from WEKA-GUI in the Package Manager and put the same package versions in my WEKA-code (I was using a maven repository for my code implementation) and the problem was solved.

Answer 2

I faced the same problem.
I think that it is about periodicity configuration (object TSLagMaker).
If you run the code from here,
it will generate the same results as the GUI. If you remove following lines, results will be different:
// add a month of the year indicator field
forecaster.getTSLagMaker().setAddMonthOfYear(true);
// add a quarter of the year indicator field
forecaster.getTSLagMaker().setAddQuarterOfYear(true);
I think that GUI detects it automatically somehow.
Hope that it helped,
Honza