private static Double evaluate(Collection<?> values, int quantile) {
List<Double> doubleValues = new ArrayList<>();
for (Object value : values) {
Double doubleValue = (Double) TypeUtil.parseOrCast(DataType.DOUBLE, value);
doubleValues.add(doubleValue);
}
double[] data = Doubles.toArray(doubleValues);
// The data must be (at least partially) ordered
Arrays.sort(data);
Percentile percentile = new Percentile();
percentile.setData(data);
return percentile.evaluate(quantile);
}
/**
* @return an array of values to split the numeric feature's values on when building candidate
* splits. When input size is <= MAX_NUMERIC_SPLITS + 1, it will return the averages between
* success values as split points. When larger, it will return MAX_NUMERIC_SPLITS approximate
* percentiles through the data.
*/
private static double[] chooseNumericSplitPoints(double[] values) {
if (values.length <= 1) {
return values;
}
if (values.length <= MAX_NUMERIC_SPLITS + 1) {
double[] splitPoints = new double[values.length - 1];
for (int i = 1; i < values.length; i++) {
splitPoints[i - 1] = (values[i] + values[i - 1]) / 2.0;
}
return splitPoints;
}
Percentile distribution = new Percentile();
distribution.setData(values);
double[] percentiles = new double[MAX_NUMERIC_SPLITS];
for (int i = 0; i < percentiles.length; i++) {
double p = 100.0 * ((i + 1.0) / (MAX_NUMERIC_SPLITS + 1.0));
percentiles[i] = distribution.evaluate(p);
}
return percentiles;
}
/**
* Returns an estimate of the <code>p</code>th percentile of the values in the <code>values</code>
* array, starting with the element in (0-based) position <code>begin</code> in the array and
* including <code>length</code> values.
*
* <p>
*
* <ul>
* <li>Returns <code>Double.NaN</code> if <code>length = 0</code>
* <li>Returns (for any value of <code>p</code>) <code>values[begin]</code> if <code>length = 1
* </code>
* <li>Throws <code>MathIllegalArgumentException</code> if <code>values</code> is null , <code>
* begin</code> or <code>length</code> is invalid, or <code>p</code> is not a valid quantile
* value (p must be greater than 0 and less than or equal to 100)
* </ul>
*
* <p>See {@link org.apache.commons.math3.stat.descriptive.rank.Percentile} for a description of
* the percentile estimation algorithm used.
*
* @param values array of input values
* @param p the percentile to compute
* @param begin the first (0-based) element to include in the computation
* @param length the number of array elements to include
* @return the percentile value
* @throws MathIllegalArgumentException if the parameters are not valid or the input array is null
*/
public static double percentile(
final double[] values, final int begin, final int length, final double p)
throws MathIllegalArgumentException {
return PERCENTILE.evaluate(values, begin, length, p);
}
