samples.cc

经网络提出了一种基于蚁群聚类算法的径向基神经网络. 利用蚁群算法的并行寻优特征和挥发系数方法的自适应更改信息量的能力,并以球面聚类的方式确定了径向基神经网络中基函数的位置, 同时通过比较隐层神经元的相

/* $Id: samples.cc,v 1.2 2006-08-09 15:20:54 jonathan Exp $
* Jonathan Ledlie, Harvard University.
* Copyright 2005.  All rights reserved.
*/

#include "sim.h"
#include "samples.h"

int PING_HISTORY_COUNT = 4;
double PING_SAMPLE_PERCENTILE = .25;
int MIN_HISTORY_COUNT = 0;

Samples::Samples()
{
	ewma = 0;
	jitterCount = 0;
	weightedError = 0;
	vec = new Point ();
	appVector = new Point ();
}

Samples::~Samples()
{
	delete vec;
	delete appVector;
}

void Samples::addSample (double sample, int myId, int yourId, 
						 Point *yourCoord, double yourWeightedError,
						 Point *yourAppCoord, int timestamp)
{
	stamp = timestamp;

	if (PING_HISTORY_COUNT == -1) 
	{
		ewma = PING_SAMPLE_PERCENTILE * sample +
			(1.-PING_SAMPLE_PERCENTILE) + ewma;
		sample = ewma;
	}

	weightedError = yourWeightedError;
	vec->assign (yourCoord);
	appVector->assign (yourAppCoord);

	// toss the sample from the front
	if (samples.size() > PING_HISTORY_COUNT)
		samples.pop_front ();
	// add this guy to the back
	samples.push_back (sample);
}

double Samples::getSample ()
{

	if (samples.size() > MIN_HISTORY_COUNT)
	{
		// sort em
		deque<double> sortedSamples (samples);
		sort (sortedSamples.begin(), sortedSamples.end());

		int percentile = (int)(samples.size() * PING_SAMPLE_PERCENTILE);
		double sample_at_percentile = sortedSamples[percentile];

		return sample_at_percentile;
	}
	else 
	{
		return -1;
	}
}

void Samples::print () 
{
	printf ("sample size %d\n", samples.size());
	for (int i = 0; i < samples.size(); i++) 
	{
		printf ("sample %d %f\n", i, samples[i]);
	}
}

ostream& operator << (ostream& os, Samples *s) 
{
	os << "[sC " << s->vec << " aC " << s->appVector 
		<< " wE " << s->weightedError << " s ";
	for (int i = 0; i < s->samples.size(); i++)
	{
		os << s->samples[i] << " ";
	} 
	os << "]";
	return os;
}