[jira] [Comment Edited] (MATH-805) Percentile calculation is very slow when input data are constants

["Benoit de Rancourt (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MATH-805?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13416943#comment-13416943 ] 

Benoit de Rancourt edited comment on MATH-805 at 7/18/12 8:42 AM:
------------------------------------------------------------------

Hello,

Here is a simple test case :

{code:title=testPercentile.java|borderStyle=solid}
/**
 * Test the Percentile calculation
 */
public static void testPercentile() {
	
	final double CONST_NUMBER = 18.;
	final double PERCENT = 5.;
	final int DISTRIBUTION_SIZE = (int) 1e5;

	double[] distribution = new double[DISTRIBUTION_SIZE];
	Percentile percentile = new Percentile(PERCENT);
	Random random = new Random(System.nanoTime());
	
	// filling the array with random number
	for (int i = 0; i < distribution.length; i++) {
		distribution[i] = random.nextDouble() * 100.;
	}
	
	System.out.println("Start the calculation with random array");
	long begin = System.currentTimeMillis();
	double result = percentile.evaluate(distribution);
	long end = System.currentTimeMillis();
	System.out.println("duration : " + (end - begin) + "ms");
	System.out.println("result : " + result);
	
	// filling the array with a constant number
	for (int i = 0; i < distribution.length; i++) {
		distribution[i] = CONST_NUMBER;
	}
	
	System.out.println("Start the calculation with constant array");
	begin = System.currentTimeMillis();
	result = percentile.evaluate(distribution);
	end = System.currentTimeMillis();
	System.out.println("duration : " + (end - begin) + "ms");
	System.out.println("result : " + result);
}
{code}

Thanks,
Benoit.
                
      was (Author: teraben):
    Hello,

Here is a simple test case :
	
/**
 * Test the Percentile calculation
 */
public static void testQuantile() {
	
	final double CONST_NUMBER = 18.;
	final double PERCENT = 5.;
	final int DISTRIBUTION_SIZE = (int) 1e5;

	double[] distribution = new double[DISTRIBUTION_SIZE];
	Percentile percentile = new Percentile(PERCENT);
	Random random = new Random(System.nanoTime());
	
	// filling the array with random number
	for (int i = 0; i < distribution.length; i++) {
		distribution[i] = random.nextDouble() * 100.;
	}
	
	System.out.println("Start the calculation with random array");
	long begin = System.currentTimeMillis();
	double result = percentile.evaluate(distribution);
	long end = System.currentTimeMillis();
	System.out.println("duration : " + (end - begin) + "ms");
	System.out.println("result : " + result);
	
	// filling the array with a constant number
	for (int i = 0; i < distribution.length; i++) {
		distribution[i] = CONST_NUMBER;
	}
	
	System.out.println("Start the calculation with constant array");
	begin = System.currentTimeMillis();
	result = percentile.evaluate(distribution);
	end = System.currentTimeMillis();
	System.out.println("duration : " + (end - begin) + "ms");
	System.out.println("result : " + result);
}

Thanks,
Benoit.
                  
> Percentile calculation is very slow when input data are constants
> -----------------------------------------------------------------
>
>                 Key: MATH-805
>                 URL: https://issues.apache.org/jira/browse/MATH-805
>             Project: Commons Math
>          Issue Type: Improvement
>    Affects Versions: 3.0
>            Reporter: Benoit de Rancourt
>            Priority: Minor
>              Labels: performance, test
>
> I use the Percentile class to calculate quantile on a big array (10^6 entries). When I have to test the performance of my code, I notice that the calculation of quantile is at least 100x slower when my data are constants (10^6 of the same nomber). Maybe the Percentile calculation can be improved for this special case.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira