typename TypeTraits<TVector>::LargerComponentType Correlation(const TVector& v1, const TVector& v2)
{
// http://docs.opencv.org/doc/tutorials/imgproc/histograms/histogram_comparison/histogram_comparison.html
// d(H_1, H_2) = \frac{\sum_i(H_1(i) - \bar{H_1})(H_2(i)-\bar{H_2})}{sqrt(var(H_1)var(H_2))}
assert(Helpers::length(v1) > 0);
assert(Helpers::length(v1) == Helpers::length(v2));
float numerator = 0.0f;
float meanV1 = Average(v1);
float meanV2 = Average(v2);
float varianceV1 = Variance(v1);
float varianceV2 = Variance(v2);
for(unsigned int i = 0; i < Helpers::length(v1); ++i)
{
numerator += (v1[i] - meanV1)*(v2[i] - meanV2);
}
float correlation = numerator/(sqrt(varianceV1 * varianceV2));
return correlation;
}
/**
* Função contruída com o objetivo de facilitar o treinamento de uma rede. Utiliza critérios
* de parada pré-definidos. O objetivo é paralizar o treinamento a partir do momento em que o
* erro médio quadrático da rede em relação às amostras para de diminuir. Recebe um parâmetro
* indicando um número mínimo de treinos, a a partir do qual se inicia a verificação da variaçao
* do erro médio quadrático. Recebe também o número de treinamentos a ser executado até que uma
* nova medição do erro seja feita. Caso a variância (porcentual) das últimas n medições seja
* menor ou igual a um determinado valor (entre 0 e 1), paraliza o treinamento.
* A função recebe ainda um conjunto de amostras (matriz de entradas/matriz de saídas), número
* de amostras contidas nas matrizes, a dimensão de cada amostra de entrada e de cada amostra de
* saída e um flag indicando se as amostras devem ser treinadas aleatoriamente ou em ordem.
*/
int BKPNeuralNet::AutoTrain( float**inMatrix, float **outMatrix, int inSize, int outSize, int nSamples,
int minTrains, int varVectorSize, float minStdDev, int numTrains, TrainType type,
float l_rate, float momentum, int* retExecutedTrains )
{
// Casos de retorno:
if( (!inMatrix) || (!outMatrix) || (inSize!=_nLayers[0]) || (_nLayers[_layers-1]!=outSize) )
return -1;
// O número de treinamentos inicial tem que ser pelo menos 0:
if( *retExecutedTrains < 0 )
*retExecutedTrains = 0;
int thisSample = -1; //< Variável auxiliar, indica a amostra a ser treinada.
// Executando os treinamentos obrigatórios:
for( int i=0 ; i<minTrains ; i++ )
{
if( type == ORDERED_TRAIN )
thisSample = (++thisSample)%nSamples;
if( type == RANDOM_TRAIN )
thisSample = RandInt(0, (nSamples-1));
Train( inSize, inMatrix[thisSample], outSize, outMatrix[thisSample], l_rate, momentum );
}
// Executando os demais treinamentos:
float* varVector = new float[varVectorSize]; //< Vetor para conter as últimas medições de erro.
int ptVarVector = 0; //< Aponta para a primeira posição vazia de varVector.
float lastVariance = (float)MAX_VALUE; //< Variâvel que mantém o valor da varirância.
float StdDev = (float)MAX_VALUE; //< Variâvel que mantém o valor do desvio-padrão.
thisSample = -1;
int nTrains=minTrains + *retExecutedTrains; //< Mantém o número de treinamentos executados.
bool varFlag = false;
while( StdDev > minStdDev )
{
if( type == ORDERED_TRAIN )
thisSample = (++thisSample)%nSamples;
if( type == RANDOM_TRAIN )
thisSample = RandInt(0, (nSamples-1));
Train( inSize, inMatrix[thisSample], outSize, outMatrix[thisSample], l_rate, momentum );
if( (nTrains%numTrains) == 0 ) //< A cada numTrains treinamentos, testa o erro:
{
float retRMS_Error = 0;
float mean = 0;
RMS_error( inMatrix, outMatrix, inSize, outSize, nSamples, &retRMS_Error );
varFlag = ShiftLeft( varVector, varVectorSize, retRMS_Error, ptVarVector );
if( varFlag == true )
{
lastVariance = Variance( varVector, varVectorSize, &mean );
StdDev = ((float)sqrt(lastVariance))/mean;
}
ptVarVector++;
}
nTrains++;
if( nTrains >= 90000 ) //< O número máximo de treinamentos será 150000.
StdDev = minStdDev;
}
*retExecutedTrains = nTrains;
return 0;
}
double ChiRand::Skewness() const
{
double mu = Mean();
double sigmaSq = Variance();
double skew = mu * (1 - 2 * sigmaSq);
skew /= std::pow(sigmaSq, 1.5);
return skew;
}
static void chanfunc_CalcStatistics (channelPtr chan)
{
double mean, std_dev, variance, rms, moment, median, mode, min, max;
int err, order, min_i, max_i, intervals;
char newnote[256];
Fmt (chanfunc.note, "");
err = MaxMin1D (chan->readings, chan->pts, &max, &max_i, &min, &min_i);
SetInputMode (chanfunc.p, STATISTICS_MIN, !err);
SetCtrlVal (chanfunc.p, STATISTICS_MIN, min);
SetInputMode (chanfunc.p, STATISTICS_MAX, !err);
SetCtrlVal (chanfunc.p, STATISTICS_MAX, max);
if (err == NoErr)
{
Fmt (chanfunc.note, "%s<Min: %f[e2p5]\n", min);
Fmt (chanfunc.note, "%s[a]<Max: %f[e2p5]\n", max);
}
err = Mean (chan->readings, chan->pts, &mean);
SetInputMode (chanfunc.p, STATISTICS_MEAN, !err);
SetCtrlVal (chanfunc.p, STATISTICS_MEAN, mean);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<Mean: %f[e2p5]\n", mean);
err = StdDev (chan->readings, chan->pts, &mean, &std_dev);
SetInputMode (chanfunc.p, STATISTICS_STDDEV, !err);
SetCtrlVal (chanfunc.p, STATISTICS_STDDEV, std_dev);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<StdDev: %f[e2p5]\n", std_dev);
err = Variance (chan->readings, chan->pts, &mean, &variance);
SetInputMode (chanfunc.p, STATISTICS_VAR, !err);
SetCtrlVal (chanfunc.p, STATISTICS_VAR, variance);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<Variance: %f[e2p5]\n", variance);
err = RMS (chan->readings, chan->pts, &rms);
SetInputMode (chanfunc.p, STATISTICS_RMS, !err);
SetCtrlVal (chanfunc.p, STATISTICS_RMS, rms);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<RMS: %f[e2p5]\n", rms);
GetCtrlVal (chanfunc.p, STATISTICS_ORDER, &order);
err = Moment (chan->readings, chan->pts, order, &moment);
SetInputMode (chanfunc.p, STATISTICS_MOMENT, !err);
SetInputMode (chanfunc.p, STATISTICS_ORDER, !err);
SetCtrlVal (chanfunc.p, STATISTICS_MOMENT, moment);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<Moment: %f[e2p5] (order: %i)\n", moment, order);
err = Median (chan->readings, chan->pts, &median);
SetInputMode (chanfunc.p, STATISTICS_MEDIAN, !err);
SetCtrlVal (chanfunc.p, STATISTICS_MEDIAN, median);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<Median: %f[e2p5]\n", median);
GetCtrlVal (chanfunc.p, STATISTICS_INTERVAL, &intervals);
err = Mode (chan->readings, chan->pts, min, max, intervals, &mode);
SetInputMode (chanfunc.p, STATISTICS_INTERVAL, !err);
SetInputMode (chanfunc.p, STATISTICS_MODE, !err);
SetCtrlVal (chanfunc.p, STATISTICS_INTERVAL, intervals);
SetCtrlVal (chanfunc.p, STATISTICS_MODE, mode);
if (err == NoErr) Fmt (chanfunc.note, "%s[a]<Mode: %f[e2p5] (intervals: %i)\n", mode, intervals);
}
double NakagamiDistribution::Skewness() const
{
double thirdMoment = lgammaShapeRatio;
thirdMoment -= 1.5 * Y.GetLogRate();
thirdMoment = (m + 0.5) * std::exp(thirdMoment);
double mean = Mean();
double variance = Variance();
return (thirdMoment - mean * (3 * variance + mean * mean)) / std::pow(variance, 1.5);
}
int main()
{
std::vector<int> a = {1,2,3,4,5,6,7};
std::vector<int> psum = PartialSum(a);
printf("%f\n", Variance(SquarePartialSum(a), PartialSum(a), 2, 4));
return 0;
}
double ChiRand::ExcessKurtosis() const
{
double mu = Mean();
double sigmaSq = Variance();
double sigma = std::sqrt(sigmaSq);
double skew = Skewness();
double kurt = 1.0 - mu * sigma * skew;
kurt /= sigmaSq;
--kurt;
return 2 * kurt;
}
double UnivariateDistribution<T>::ThirdMoment() const
{
double mean = Mean();
double variance = Variance();
double skewness = Skewness();
double moment = skewness * std::sqrt(variance) * variance;
moment += mean * mean * mean;
moment += 3 * mean * variance;
return moment;
}
void ParallelPlane::UpdateText(){
float var = Variance();
QString var_text = (var<0)?"NA":QString("Variance:%1").arg(var);
text_->setPosition(osg::Vec3(0.02,0.1,0.01f));
QString txt = QString("XAxis: %1\nYAxis: %2\n%3").arg(db_->get_header(axes_[0])).arg(db_->get_header(axes_[1])).arg(var_text);
printf("Text: %s\n",txt.toStdString().c_str());
text_->setText(txt.toStdString());
}
void RunningStats::Print(FILE * pFile, const char * header) const
{
fprintf (pFile, "\n%s\n", header);
fprintf (pFile, "NumDataValues: %ld\n", NumDataValues());
fprintf (pFile, "Mean: %f\n", Mean());
fprintf (pFile, "Variance: %f\n", Variance());
fprintf (pFile, "StandardDeviation: %f\n", StandardDeviation());
fprintf (pFile, "Skewness: %f\n", Skewness());
fprintf (pFile, "Kurtosis: %f\n", Kurtosis());
fprintf (pFile, "Maximum: %f\n", Maximum());
fprintf (pFile, "Minimum: %f\n", Minimum());
return;
}
double UnivariateDistribution<T>::FourthMoment() const
{
double mean = Mean();
double variance = Variance();
double moment3 = ThirdMoment();
double kurtosis = Kurtosis();
double meanSq = mean * mean;
double moment = kurtosis * variance * variance;
moment -= 6 * meanSq * variance;
moment -= 3 * meanSq * meanSq;
moment += 4 * mean * moment3;
return moment;
}
double UnivariateDistribution<T>::Skewness() const
{
double var = Variance();
if (!std::isfinite(var))
return NAN;
double mu = Mean(); /// var is finite, so is mu
double sum = ExpectedValue([this, mu] (double x)
{
double xmmu = x - mu;
double skewness = xmmu * xmmu * xmmu;
return skewness;
}, this->MinValue(), this->MaxValue());
return sum / std::pow(var, 1.5);
}
double UnivariateDistribution<T>::ExcessKurtosis() const
{
double var = Variance();
if (!std::isfinite(var))
return NAN;
double mu = Mean(); /// var is finite, so is mu
double sum = ExpectedValue([this, mu] (double x)
{
double xmmu = x - mu;
double kurtosisSqrt = xmmu * xmmu;
double kurtosis = kurtosisSqrt * kurtosisSqrt;
return kurtosis;
}, this->MinValue(), this->MaxValue());
return sum / (var * var) - 3;
}
/// <summary>
/// Computes the standard deviation for a given history.
/// </summary>
/// <returns>
/// The standard deviation.
/// </returns>
double HillClimbing::MeasuredHistory::StandardDeviation()
{
return sqrt(Variance());
}
double RunningStats::StandardDeviation() const
{
return sqrt( Variance() );
}
void
JPlotLinearFit::LinearLSQ1()
{
const JPlotDataBase* data = GetDataToFit();
J2DDataPoint point;
const JSize count = data->GetElementCount();
JArray<JFloat> weight;
JArray<JFloat> sigma;
JSize i;
JFloat vx, vy;
Variance(&vx,&vy);
JFloat resize = vy/vx;
JFloat num = 0;
JFloat avgx = 0;
for (i = 1; i <= count; i++)
{
if (GetDataElement(i, &point))
{
JFloat sy = point.yerr;
JFloat sx = point.xerr;
if (sy == 0)
{
sy = 1;
}
JFloat s = 0;
if (!itsXIsLog)
{
if (!itsYIsLog)
{
s = sqrt(sy * sy + resize * resize * sx * sx);
}
else
{
s = sqrt(sy * sy + resize * resize * point.y * point.y * sx * sx);
}
}
else
{
// do for power law;
}
sigma.AppendElement(s);
JFloat w = 1/(s*s);
weight.AppendElement(w);
num += w;
avgx += w * point.x;
itsRealCount++;
}
}
avgx /= num;
JArray<JFloat> t;
JFloat stt = 0;
JFloat b = 0;
JSize counter = 1;
for (i = 1; i <= count; i++)
{
if (GetDataElement(i, &point))
{
JFloat tTemp = (point.x - avgx)/(sigma.GetElement(counter));
t.AppendElement(tTemp);
stt += tTemp * tTemp;
b += tTemp * point.y / sigma.GetElement(counter);
counter++;
}
}
b /= stt;
JFloat a = 0;
JFloat aerr = 0;
counter = 1;
for (i = 1; i <= count; i++)
{
if (GetDataElement(i, &point))
{
JFloat w = weight.GetElement(counter);
a += w * (point.y - b * point.x);
aerr += w * w * point.x * point.x;
counter++;
}
}
a /= num;
aerr /= (num * stt);
aerr += 1;
aerr /= num;
aerr = sqrt(aerr);
JFloat berr = sqrt(1.0/stt);
JFloat c = 0;
JBoolean sytest = kJTrue;
counter = 1;
for (i = 1; i <= count; i++)
{
if (GetDataElement(i, &point))
{
JFloat temp = (point.y - a - b * point.x)/(sigma.GetElement(counter));
c += temp * temp;
if (sigma.GetElement(counter) != 1)
{
sytest = kJFalse;
}
counter++;
}
}
if (sytest)
{
JFloat sig = sqrt(c/count);
aerr = aerr * sig;
berr = berr * sig;
}
itsAParameter = a;
itsAErrParameter = aerr;
itsBParameter = b;
itsBErrParameter = berr;
itsChi2 = c;
}
double StandardDeviation() const
{
return sqrt( Variance() );
}
double UnivariateDistribution<T>::SecondMoment() const
{
double mean = Mean();
return mean * mean + Variance();
}
double DataSource::StdDev(Getdatafun getdata, double avg) {
return sqrt(Variance(getdata, avg));
}
void System::runMonteCarlo()
{
int i, j, k, NOA;
int alpha_max = Alpha.n_elem;
int beta_max = Beta.n_elem;
double I, I2, dx; // total energy integral
double T, T2, dt; // kinetic energy integral
double V, V2, dv; // potential energy integral
bool Accepted;
for (i=0; i<alpha_max; i++) // LOOP OVER ALPHA VALUES
{
Wavefunction->setAlpha(i);
TypeHamiltonian->getWavefunction()->setAlpha(i);
for (j=0; j<beta_max; j++) // LOOP OVER BETA VALUES
{
Wavefunction->setBeta(j);
TypeHamiltonian->getWavefunction()->setBeta(j);
dx = I = I2 = NOA = 0;
dt = T = T2 = 0;
dv = V = V2 = 0;
// BRUTE FORCE METROPOLIS:
for (k=0; k<NumberOfCycles; k++)
{
Accepted = newStepMetropolis(); // NEW STEP: ACCEPTED OR REFUSED
if (Accepted)
{
// LOCAL ENERGY
dx = TypeHamiltonian->evaluateLocalEnergy(OldPosition);
I += dx;
I2 += dx*dx;
// LOCAL KINETIC ENERGY
dt = TypeHamiltonian->getKineticEnergy();
T += dt;
T2 += dt*dt;
// LOCAL POTENTIAL ENERGY
dv = TypeHamiltonian->getPotentialEnergy();
V += dv;
V2 += dv*dv;
NOA++;
}
else
{
I += dx;
I2 += dx*dx;
T += dt;
T2 += dt*dt;
V += dv;
V2 += dv*dv;
}
}
NumberOfAcceptedSteps(i,j) = NOA;
Energy(i,j) = I/double(NumberOfCycles);
EnergySquared(i,j) = I2/double(NumberOfCycles);
Variance(i,j) = (EnergySquared(i,j) - Energy(i,j)*Energy(i,j));
KineticEnergy(i,j) = T/double(NumberOfCycles);
KineticEnergySquared(i,j) = T2/double(NumberOfCycles);
PotentialEnergy(i,j) = V/double(NumberOfCycles);
PotentialEnergySquared(i,j) = V2/double(NumberOfCycles);
}
}
}
void System::importanceSampling()
{
int NOA; // Number Of Accepted steps
double I, I2, dx; // total energy integral
double T, T2, dt; // kinetic energy integral
double V, V2, dv; // potential energy integral
int a,b,c,i,j; // loop variables: a->alpha, b->beta, c->cycles, i->particle, j-> dimension
int amax = Alpha.n_elem; // number of total alpha values
int bmax = Beta.n_elem; // number of total beta values
double wf_new, wf_old;
double greensfunction;
double D = 0.5; // diffusion constant
for (a=0; a<amax; a++) // LOOP OVER ALPHA VALUES
{
Wavefunction->setAlpha(a);
TypeHamiltonian->getWavefunction()->setAlpha(a);
for (b=0; b<bmax; b++) // LOOP OVER BETA VALUES
{
Wavefunction->setBeta(b);
TypeHamiltonian->getWavefunction()->setBeta(b);
dx = I = I2 = NOA = 0;
dt = T = T2 = 0;
dv = V = V2 = 0;
// IMPORTANCE SAMPLING:
for (c=0; c<NumberOfCycles; c++)
{
dx = 0;
for (i=0; i<NumberOfParticles; i++)
{
// Taking a new, random step, moving one particle only:
NewPosition = OldPosition;
NewPosition.row(i) = OldPosition.row(i)+Rnd->nextGauss(0,sqrt(StepLength))+QuantumForceOld.row(i)*StepLength*D;
wf_new = Wavefunction->evaluateWavefunction(NewPosition);
quantumForce(NewPosition,QuantumForceNew,wf_new);
// Metropolis-Hastings algorithm:
greensfunction = 0.0;
for(j=0;j<NumberOfDimensions;j++)
{
greensfunction += 0.5*(QuantumForceOld(i,j) + QuantumForceNew(i,j))*(D*StepLength*0.5*(QuantumForceOld(i,j)-QuantumForceNew(i,j)) - NewPosition(i,j) + OldPosition(i,j));
}
greensfunction = exp(greensfunction);
// Metropolis test:
if (ran0(&RandomSeed) <= greensfunction*wf_new*wf_new/(wf_old*wf_old))
{
OldPosition.row(i) = NewPosition.row(i);
QuantumForceOld.row(i) = QuantumForceNew.row(i);
Wavefunction->setOldWavefunction(wf_new);
wf_old = wf_new;
}
}
// Updating integral:
// LOCAL ENERGY
dx = TypeHamiltonian->evaluateLocalEnergy(OldPosition);
I += dx;
I2 += dx*dx;
// LOCAL KINETIC ENERGY
dt = TypeHamiltonian->getKineticEnergy();
T += dt;
T2 += dt*dt;
// LOCAL POTENTIAL ENERGY
dv = TypeHamiltonian->getPotentialEnergy();
V += dv;
V2 += dv*dv;
NOA++;
}
NumberOfAcceptedSteps(a,b) = NOA;
Energy(a,b) = I/double(NumberOfCycles);
EnergySquared(a,b) = I2/double(NumberOfCycles);
Variance(a,b) = (EnergySquared(a,b) - Energy(a,b)*Energy(a,b));
KineticEnergy(a,b) = T/double(NumberOfCycles);
KineticEnergySquared(a,b) = T2/double(NumberOfCycles);
PotentialEnergy(a,b) = V/double(NumberOfCycles);
PotentialEnergySquared(a,b) = V2/double(NumberOfCycles);
//AvgDistance = 0;
}
}
}
double CAfficheMesPol::Correlation(int par1,int par2)
{
return (Covariance(par1,par2)/(pow(Variance(par1)*Variance(par2),0.5)));
}
double DataSource::StdDev(Getdatafun getdata, double avg) {
double var = Variance(getdata, avg);
return IsNull(var) ? Null : sqrt(var);
}
/* Calcula o desvio padrao de um vetor de valores double.
Parametro 1: vetor tipo double.
Parametro 2: tamanho do vetor. */
double StandardDeviation(double *vec, int length)
{
return sqrt(Variance(vec, length));
}
/// <summary>
/// Computes the mean of variances for a given history.
/// </summary>
/// <returns>
/// The mean of variances.
/// </returns>
double HillClimbing::MeasuredHistory::VarianceMean()
{
return Variance() / Count();
}
