/* (non-Javadoc)
* @see org.apache.hadoop.mapreduce.Reducer#setup(org.apache.hadoop.mapreduce.Reducer.Context)
*/
protected void setup(Context context) throws IOException, InterruptedException {
Configuration config = context.getConfiguration();
if (config.getBoolean("debug.on", false)) {
LOG.setLevel(Level.DEBUG);
System.out.println("in debug mode");
}
fieldDelim = config.get("field.delim", ",");
topMatchCount = config.getInt("top.match.count", 10);
isValidationMode = config.getBoolean("validation.mode", true);
kernelFunction = config.get("kernel.function", "none");
kernelParam = config.getInt("kernel.param", -1);
classCondtionWeighted = config.getBoolean("class.condtion.weighted", false);
neighborhood = new Neighborhood(kernelFunction, kernelParam, classCondtionWeighted);
outputClassDistr = config.getBoolean("output.class.distr", false);
inverseDistanceWeighted = config.getBoolean("inverse.distance.weighted", false);
// regression
String predictionMode = config.get("prediction.mode", "classification");
if (predictionMode.equals("regression")) {
neighborhood.withPredictionMode(PredictionMode.Regression);
String regressionMethod = config.get("regression.method", "average");
regressionMethod = WordUtils.capitalize(regressionMethod);
neighborhood.withRegressionMethod(RegressionMethod.valueOf(regressionMethod));
}
// decision threshold for classification
decisionThreshold = Double.parseDouble(config.get("decision.threshold", "-1.0"));
if (decisionThreshold > 0 && neighborhood.IsInClassificationMode()) {
String[] classAttrValues = config.get("class.attribute.values").split(",");
posClassAttrValue = classAttrValues[0];
negClassAttrValue = classAttrValues[1];
neighborhood.withDecisionThreshold(decisionThreshold).withPositiveClass(posClassAttrValue);
}
// using cost based arbitrator for classification
useCostBasedClassifier = config.getBoolean("use.cost.based.classifier", false);
if (useCostBasedClassifier && neighborhood.IsInClassificationMode()) {
if (null == posClassAttrValue) {
String[] classAttrValues = config.get("class.attribute.values").split(",");
posClassAttrValue = classAttrValues[0];
negClassAttrValue = classAttrValues[1];
}
int[] missclassificationCost =
Utility.intArrayFromString(config.get("misclassification.cost"));
falsePosCost = missclassificationCost[0];
falseNegCost = missclassificationCost[1];
costBasedArbitrator =
new CostBasedArbitrator(
negClassAttrValue, posClassAttrValue, falseNegCost, falsePosCost);
}
// confusion matrix for classification validation
if (isValidationMode) {
if (neighborhood.IsInClassificationMode()) {
InputStream fs =
Utility.getFileStream(context.getConfiguration(), "feature.schema.file.path");
ObjectMapper mapper = new ObjectMapper();
schema = mapper.readValue(fs, FeatureSchema.class);
classAttrField = schema.findClassAttrField();
List<String> cardinality = classAttrField.getCardinality();
predictingClasses = new String[2];
predictingClasses[0] = cardinality.get(0);
predictingClasses[1] = cardinality.get(1);
confMatrix = new ConfusionMatrix(predictingClasses[0], predictingClasses[1]);
}
}
LOG.debug(
"classCondtionWeighted:"
+ classCondtionWeighted
+ "outputClassDistr:"
+ outputClassDistr);
}
