Load bayesian network based on BIF-file WEKA - weka

Question

I want to load a bayesian network that is stored in a BIF-file in WEKA to validate it against a specific test-set.
However it keeps crashing with a null-pointer because m_MissinValuesFilter is null.
How do I set this filter correctly?
My current code looks as follows:
BayesNet network = new BayesNet()
BIFReader reader = new BIFReader()
network reader.processFile(path)
Evaluation eval = new Evaluation(testData)
eval.evaluatemodel(network,testData)

Answer 1

Despite loading the network from a BIF XML file, you still need to call the buildClassifier method to initialize all the data structures.
Here is a minimal code example:
import weka.classifiers.Evaluation;
import weka.classifiers.bayes.BayesNet;
import weka.classifiers.bayes.net.search.fixed.FromFile;
import weka.core.Instances;
import weka.core.converters.ConverterUtils;
public class Bif {
public static void main(String[] args) throws Exception {
// load data
Instances data = ConverterUtils.DataSource.read("/somewhere/data.arff");
data.setClassIndex(data.numAttributes() - 1);
// configure classifier
FromFile fromFile = new FromFile();
fromFile.setBIFFile("/somewhere/model.bif");
BayesNet bayesNet = new BayesNet();
bayesNet.setSearchAlgorithm(fromFile);
// build classifier to initialize data structures
// NB: this won't change the model itself, as it is fixed
bayesNet.buildClassifier(data);
// evaluate it
Evaluation eval = new Evaluation(data);
eval.evaluateModel(bayesNet, data);
System.out.println(eval.toSummaryString());
// you can output the BIF XML like this
//System.out.println(bayesNet.graph());
}
}
This approach will, however, only use the structure of the network stored in the BIF file and not the CPDs. The training process will reinitialize the CPDs.
If you also want to use the CPDs from the BIF file, you need to use this slightly hacky approach:
import weka.classifiers.Evaluation;
import weka.classifiers.bayes.net.BIFReader;
import weka.core.Instances;
import weka.core.converters.ConverterUtils;
public class Bif {
public static void main(String[] args) throws Exception {
Instances test = ConverterUtils.DataSource.read("/somewhere/data.arff");
test.setClassIndex(test.numAttributes() - 1);
String bifFile = "/somewhere/model.bif";
BIFReader reader = new BIFReader();
// initialize internal filters etc
reader.buildClassifier(test);
// override data structures using BIF file
reader.processFile(bifFile);
// output graph to confirm correct model
//System.out.println(reader.graph());
// evaluate the model against the test set
Evaluation eval = new Evaluation(test);
eval.evaluateModel(reader, test);
System.out.println(eval.toSummaryString());
}
}