All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.seleniumhq.jetty7.util.statistic.SampleStatistic Maven / Gradle / Ivy

There is a newer version: 9.4.12.v20180830.1
Show newest version
// ========================================================================
// Copyright (c) 2009-2009 Mort Bay Consulting Pty. Ltd.
// ------------------------------------------------------------------------
// All rights reserved. This program and the accompanying materials
// are made available under the terms of the Eclipse Public License v1.0
// and Apache License v2.0 which accompanies this distribution.
// The Eclipse Public License is available at 
// http://www.eclipse.org/legal/epl-v10.html
// The Apache License v2.0 is available at
// http://www.opensource.org/licenses/apache2.0.php
// You may elect to redistribute this code under either of these licenses. 
// ========================================================================

package org.eclipse.jetty.util.statistic;

import java.util.concurrent.atomic.AtomicLong;


/* ------------------------------------------------------------ */
/**
 * SampledStatistics
 * 

* Provides max, total, mean, count, variance, and standard * deviation of continuous sequence of samples. *

* Calculates estimates of mean, variance, and standard deviation * characteristics of a sample using a non synchronized * approximation of the on-line algorithm presented * in Donald Knuth's Art of Computer Programming, Volume 2, * Seminumerical Algorithms, 3rd edition, page 232, * Boston: Addison-Wesley. that cites a 1962 paper by B.P. Welford * that can be found by following the link http://www.jstor.org/pss/1266577 *

* This algorithm is also described in Wikipedia at * http://en.wikipedia.org/w/index.php?title=Algorithms_for_calculating_variance§ion=4#On-line_algorithm */ public class SampleStatistic { protected final AtomicLong _max = new AtomicLong(); protected final AtomicLong _total = new AtomicLong(); protected final AtomicLong _count = new AtomicLong(); protected final AtomicLong _totalVariance100 = new AtomicLong(); public void reset() { _max.set(0); _total.set(0); _count.set(0); _totalVariance100.set(0); } public void set(final long sample) { long total = _total.addAndGet(sample); long count = _count.incrementAndGet(); if (count>1) { long mean10 = total*10/count; long delta10 = sample*10 - mean10; _totalVariance100.addAndGet(delta10*delta10); } long oldMax = _max.get(); while (sample > oldMax) { if (_max.compareAndSet(oldMax, sample)) break; oldMax = _max.get(); } } /* ------------------------------------------------------------ */ /** * @return the max value */ public long getMax() { return _max.get(); } /* ------------------------------------------------------------ */ public long getTotal() { return _total.get(); } /* ------------------------------------------------------------ */ public long getCount() { return _count.get(); } /* ------------------------------------------------------------ */ public double getMean() { return (double)_total.get()/_count.get(); } /* ------------------------------------------------------------ */ public double getVariance() { final long variance100 = _totalVariance100.get(); final long count = _count.get(); return count>1?((double)variance100)/100.0/(count-1):0.0; } /* ------------------------------------------------------------ */ public double getStdDev() { return Math.sqrt(getVariance()); } }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy