/**
* Calculate the average value but removing outliers based on Median Absolute Deviation with a
* fixed threshold of 3.0 TODO make the threshold be the first parameter in the list or make a
* smarter analysis of method to use, not just MAD, based on the number of data in the set,
* variance, etc
*/
@SuppressWarnings("unchecked")
@Override
public void run(Stack stack) throws ParseException {
checkStack(stack);
if (curNumberOfParameters < 1) {
throw new ParseException("No arguments for Average MAD - median average deviation");
}
// Get the first parameter that has to be the threshold
// Double threshold = (Double) stack.pop();
Double threshold = 3.0;
Object param;
ArrayList<Double> orginalValues = new ArrayList<Double>();
int i = 0;
while (i < (curNumberOfParameters)) {
param = stack.pop();
if (!(supportNull && param instanceof Null)) {
orginalValues.add((Double) param);
}
i++;
}
// calculate median
Double median = median(orginalValues);
ArrayList<Double> absMedianValues = new ArrayList<Double>();
// calculate mad by abs(median - value) for each
for (Double value : orginalValues) {
absMedianValues.add(Math.abs(value - median));
}
Double mad = median(absMedianValues);
// select only values in the range of the mad*threshold => removing the
// outliers
int count = 0;
int index = 0;
Double sum = new Double(0);
while (index < absMedianValues.size()) {
if (absMedianValues.get(index) < mad * threshold) {
sum += orginalValues.get(index);
count++;
}
index++;
}
if (index > 0) {
stack.push(sum / count);
} else {
stack.push(new Null());
}
}
private Double median(ArrayList<Double> median1) {
// Calculate the median
Double medianValue = null;
Object[] objectArr = median1.toArray();
Double[] median = Arrays.copyOf(objectArr, objectArr.length, Double[].class);
// get the numbers into a sorted order
Arrays.sort(median);
if ((median.length & 1) == 0) {
// even number of parameters
medianValue = (median[median.length / 2 - 1] + median[median.length / 2]) / 2;
} else {
// odd number of parameters
medianValue = median[median.length / 2];
}
return medianValue;
}
