1
0
mirror of https://github.com/aportelli/LatAnalyze.git synced 2024-11-10 00:45:36 +00:00
LatAnalyze/lib/Statistics/XYStatData.cpp

611 lines
17 KiB
C++

/*
* XYStatData.cpp, part of LatAnalyze 3
*
* Copyright (C) 2013 - 2016 Antonin Portelli
*
* LatAnalyze 3 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* LatAnalyze 3 is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with LatAnalyze 3. If not, see <http://www.gnu.org/licenses/>.
*/
#include <LatAnalyze/Statistics/XYStatData.hpp>
#include <LatAnalyze/includes.hpp>
#include <LatAnalyze/Core/Math.hpp>
using namespace std;
using namespace Latan;
static constexpr double maxXsiDev = 10.;
/******************************************************************************
* FitResult implementation *
******************************************************************************/
// access //////////////////////////////////////////////////////////////////////
double FitResult::getChi2(void) const
{
return chi2_;
}
double FitResult::getChi2PerDof(void) const
{
return chi2_/getNDof();
}
double FitResult::getNDof(void) const
{
return static_cast<double>(nDof_);
}
Index FitResult::getNPar(void) const
{
return nPar_;
}
double FitResult::getPValue(void) const
{
return Math::chi2PValue(getChi2(), getNDof());;
}
const DoubleFunction & FitResult::getModel(const Index j) const
{
return model_[j];
}
// IO //////////////////////////////////////////////////////////////////////////
void FitResult::print(const bool printXsi, ostream &out) const
{
char buf[256];
Index pMax = printXsi ? size() : nPar_;
sprintf(buf, "chi^2/dof= %.1e/%d= %.2e -- p-value= %.2e", getChi2(),
static_cast<int>(getNDof()), getChi2PerDof(), getPValue());
out << buf << endl;
for (Index p = 0; p < pMax; ++p)
{
sprintf(buf, "%8s= %e", parName_[p].c_str(), (*this)(p));
out << buf << endl;
}
}
/******************************************************************************
* XYStatData implementation *
******************************************************************************/
// data access /////////////////////////////////////////////////////////////////
double & XYStatData::x(const Index r, const Index i)
{
checkXIndex(r, i);
scheduleXMapInit();
scheduleChi2DataVecInit();
return xData_[i](r);
}
const double & XYStatData::x(const Index r, const Index i) const
{
checkXIndex(r, i);
return xData_[i](r);
}
const DVec & XYStatData::x(const Index k) const
{
checkDataIndex(k);
updateXMap();
return xMap_.at(k);
}
double & XYStatData::y(const Index k, const Index j)
{
checkYDim(j);
if (!pointExists(k, j))
{
registerDataPoint(k, j);
resizeVarMat();
}
scheduleXMapInit();
scheduleChi2DataVecInit();
return yData_[j][k];
}
const double & XYStatData::y(const Index k, const Index j) const
{
checkPoint(k, j);
return yData_[j].at(k);
}
void XYStatData::setXXVar(const Index i1, const Index i2, const DMat &m)
{
checkXDim(i1);
checkXDim(i2);
checkVarMat(m, xxVar_(i1, i2));
xxVar_(i1, i2) = m;
if (i1 != i2)
{
xxVar_(i2, i1) = m.transpose();
}
scheduleFitVarMatInit();
}
void XYStatData::setYYVar(const Index j1, const Index j2, const DMat &m)
{
checkYDim(j1);
checkYDim(j2);
checkVarMat(m, yyVar_(j1, j2));
yyVar_(j1, j2) = m;
if (j1 != j2)
{
yyVar_(j2, j1) = m.transpose();
}
scheduleFitVarMatInit();
}
void XYStatData::setXYVar(const Index i, const Index j, const DMat &m)
{
checkXDim(i);
checkYDim(j);
checkVarMat(m, xyVar_(i, j));
xyVar_(i, j) = m;
scheduleFitVarMatInit();
}
void XYStatData::setXError(const Index i, const DVec &err)
{
checkXDim(i);
checkErrVec(err, xxVar_(i, i));
xxVar_(i, i).diagonal() = err.cwiseProduct(err);
scheduleFitVarMatInit();
}
void XYStatData::setYError(const Index j, const DVec &err)
{
checkXDim(j);
checkErrVec(err, yyVar_(j, j));
yyVar_(j, j).diagonal() = err.cwiseProduct(err);
scheduleFitVarMatInit();
}
const DMat & XYStatData::getXXVar(const Index i1, const Index i2) const
{
checkXDim(i1);
checkXDim(i2);
return xxVar_(i1, i2);
}
const DMat & XYStatData::getYYVar(const Index j1, const Index j2) const
{
checkYDim(j1);
checkYDim(j2);
return yyVar_(j1, j2);
}
const DMat & XYStatData::getXYVar(const Index i, const Index j) const
{
checkXDim(i);
checkYDim(j);
return xyVar_(i, j);
}
DVec XYStatData::getXError(const Index i) const
{
checkXDim(i);
return xxVar_(i, i).diagonal().cwiseSqrt();
}
DVec XYStatData::getYError(const Index j) const
{
checkXDim(j);
return yyVar_(j, j).diagonal().cwiseSqrt();
}
DMat XYStatData::getTable(const Index i, const Index j) const
{
checkXDim(i);
checkYDim(j);
DMat table(getYSize(j), 4);
Index row = 0;
for (auto &p: yData_[j])
{
Index k = p.first;
Index r = dataCoord(k)[i];
table(row, 0) = x(k)(i);
table(row, 2) = p.second;
table(row, 1) = xxVar_(i, i).diagonal().cwiseSqrt()(r);
table(row, 3) = yyVar_(j, j).diagonal().cwiseSqrt()(row);
row++;
}
return table;
}
// get total fit variance matrix ///////////////////////////////////////////////
const DMat & XYStatData::getFitVarMat(void)
{
updateFitVarMat();
return fitVar_;
}
const DMat & XYStatData::getFitVarMatPInv(void)
{
updateFitVarMat();
return fitVarInv_;
}
// fit /////////////////////////////////////////////////////////////////////////
FitResult XYStatData::fit(vector<Minimizer *> &minimizer, const DVec &init,
const vector<const DoubleModel *> &v)
{
// check model consistency
checkModelVec(v);
// buffering
updateLayout();
updateFitVarMat();
updateChi2DataVec();
// get number of parameters
Index nPar = v[0]->getNPar();
Index nXDim = getNXDim();
Index totalNPar = nPar + layout.totalXSize;
// chi^2 functions
auto corrChi2Func = [this, nPar, nXDim, totalNPar, &v](const double *x)->double
{
ConstMap<DVec> p(x, totalNPar);
updateChi2ModVec(p, v, nPar, nXDim);
chi2Vec_ = (chi2ModVec_ - chi2DataVec_);
return chi2Vec_.dot(fitVarInv_*chi2Vec_);
};
DoubleFunction corrChi2(corrChi2Func, totalNPar);
auto uncorrChi2Func = [this, nPar, nXDim, totalNPar, &v](const double *x)->double
{
ConstMap<DVec> p(x, totalNPar);
updateChi2ModVec(p, v, nPar, nXDim);
chi2Vec_ = (chi2ModVec_ - chi2DataVec_);
return chi2Vec_.dot(chi2Vec_.cwiseQuotient(fitVar_.diagonal()));
};
DoubleFunction uncorrChi2(uncorrChi2Func, totalNPar);
DoubleFunction &chi2 = hasCorrelations() ? corrChi2 : uncorrChi2;
for (Index p = 0; p < nPar; ++p)
{
chi2.varName().setName(p, v[0]->parName().getName(p));
}
for (Index p = 0; p < totalNPar - nPar; ++p)
{
chi2.varName().setName(p + nPar, "xsi_" + strFrom(p));
}
// minimization
FitResult result;
DVec totalInit(totalNPar);
//// set total init vector
totalInit.segment(0, nPar) = init;
totalInit.segment(nPar, layout.totalXSize) =
chi2DataVec_.segment(layout.totalYSize, layout.totalXSize);
for (auto &m: minimizer)
{
m->setInit(totalInit);
if (m->supportLimits())
{
//// do not allow more than maxXsiDev std. deviations on the x-axis
for (Index p = nPar; p < totalNPar; ++p)
{
double err;
err = sqrt(fitVar_.diagonal()(layout.totalYSize + p - nPar));
m->useLowLimit(p);
m->useHighLimit(p);
m->setLowLimit(p, totalInit(p) - maxXsiDev*err);
m->setHighLimit(p, totalInit(p) + maxXsiDev*err);
}
}
//// minimize and store results
result = (*m)(chi2);
totalInit = result;
}
result.chi2_ = chi2(result);
result.nPar_ = nPar;
result.nDof_ = layout.totalYSize - nPar;
result.model_.resize(v.size());
for (unsigned int j = 0; j < v.size(); ++j)
{
result.model_[j] = v[j]->fixPar(result);
}
for (Index p = 0; p < totalNPar; ++p)
{
result.parName_.push_back(chi2.varName().getName(p));
}
return result;
}
FitResult XYStatData::fit(Minimizer &minimizer, const DVec &init,
const vector<const DoubleModel *> &v)
{
vector<Minimizer *> mv{&minimizer};
return fit(mv, init, v);
}
// residuals ///////////////////////////////////////////////////////////////////
XYStatData XYStatData::getResiduals(const FitResult &fit)
{
XYStatData res(*this);
for (Index j = 0; j < getNYDim(); ++j)
{
const DoubleFunction &f = fit.getModel(j);
for (auto &p: yData_[j])
{
res.y(p.first, j) -= f(x(p.first));
}
}
return res;
}
XYStatData XYStatData::getPartialResiduals(const FitResult &fit,
const DVec &ref, const Index i)
{
XYStatData res(*this);
DVec buf(ref);
for (Index j = 0; j < res.getNYDim(); ++j)
{
const DoubleFunction &f = fit.getModel(j);
for (auto &p: yData_[j])
{
buf(i) = x(p.first)(i);
res.y(p.first, j) -= f(x(p.first)) - f(buf);
}
}
return res;
}
// create data /////////////////////////////////////////////////////////////////
void XYStatData::createXData(const std::string name __dumb, const Index nData)
{
xData_.push_back(DVec::Zero(nData));
xBuf_.resize(xData_.size());
resizeVarMat();
}
void XYStatData::createYData(const std::string name __dumb)
{
yData_.push_back(map<Index, double>());
resizeVarMat();
}
void XYStatData::resizeVarMat(void)
{
xxVar_.conservativeResize(getNXDim(), getNXDim());
for (Index i1 = 0; i1 < getNXDim(); ++i1)
for (Index i2 = 0; i2 < getNXDim(); ++i2)
{
xxVar_(i1, i2).conservativeResize(getXSize(i1), getXSize(i2));
}
yyVar_.conservativeResize(getNYDim(), getNYDim());
for (Index j1 = 0; j1 < getNYDim(); ++j1)
for (Index j2 = 0; j2 < getNYDim(); ++j2)
{
yyVar_(j1, j2).conservativeResize(getYSize(j1), getYSize(j2));
}
xyVar_.conservativeResize(getNXDim(), getNYDim());
for (Index i = 0; i < getNXDim(); ++i)
for (Index j = 0; j < getNYDim(); ++j)
{
xyVar_(i, j).conservativeResize(getXSize(i), getYSize(j));
}
scheduleFitVarMatInit();
}
// schedule buffer computation /////////////////////////////////////////////////
void XYStatData::scheduleXMapInit(void)
{
initXMap_ = true;
}
void XYStatData::scheduleChi2DataVecInit(void)
{
initChi2DataVec_ = true;
}
// buffer total fit variance matrix ////////////////////////////////////////////
void XYStatData::updateFitVarMat(void)
{
if (initVarMat())
{
updateLayout();
DMat &v = fitVar_;
Index roffs, coffs;
v.resize(layout.totalSize, layout.totalSize);
roffs = layout.totalYSize;
for (Index ifit1 = 0; ifit1 < layout.nXFitDim; ++ifit1)
{
coffs = layout.totalYSize;
for (Index ifit2 = 0; ifit2 < layout.nXFitDim; ++ifit2)
{
for (Index rfit1 = 0; rfit1 < layout.xSize[ifit1]; ++rfit1)
for (Index rfit2 = 0; rfit2 < layout.xSize[ifit2]; ++rfit2)
{
Index i1, i2, r1, r2;
i1 = layout.xDim[ifit1];
i2 = layout.xDim[ifit2];
r1 = layout.x[ifit1][rfit1];
r2 = layout.x[ifit2][rfit2];
v(roffs+rfit1, coffs+rfit2) = xxVar_(i1, i2)(r1, r2);
v(coffs+rfit2, roffs+rfit1) = v(roffs+rfit1, coffs+rfit2);
}
coffs += layout.xSize[ifit2];
}
roffs += layout.xSize[ifit1];
}
roffs = 0;
for (Index jfit1 = 0; jfit1 < layout.nYFitDim; ++jfit1)
{
coffs = 0;
for (Index jfit2 = 0; jfit2 < layout.nYFitDim; ++jfit2)
{
for (Index sfit1 = 0; sfit1 < layout.ySize[jfit1]; ++sfit1)
for (Index sfit2 = 0; sfit2 < layout.ySize[jfit2]; ++sfit2)
{
Index j1, j2, s1, s2;
j1 = layout.yDim[jfit1];
j2 = layout.yDim[jfit2];
s1 = layout.y[jfit1][sfit1];
s2 = layout.y[jfit2][sfit2];
v(roffs+sfit1, coffs+sfit2) = yyVar_(j1, j2)(s1, s2);
v(coffs+sfit2, roffs+sfit1) = v(roffs+sfit1, coffs+sfit2);
}
coffs += layout.ySize[jfit2];
}
roffs += layout.ySize[jfit1];
}
roffs = layout.totalYSize;
for (Index ifit = 0; ifit < layout.nXFitDim; ++ifit)
{
coffs = 0;
for (Index jfit = 0; jfit < layout.nYFitDim; ++jfit)
{
for (Index rfit = 0; rfit < layout.xSize[ifit]; ++rfit)
for (Index sfit = 0; sfit < layout.ySize[jfit]; ++sfit)
{
Index i, j, r, s;
i = layout.xDim[ifit];
j = layout.yDim[jfit];
r = layout.x[ifit][rfit];
s = layout.y[jfit][sfit];
v(roffs+rfit, coffs+sfit) = xyVar_(i, j)(r, s);
v(coffs+sfit, roffs+rfit) = v(roffs+rfit, coffs+sfit);
}
coffs += layout.ySize[jfit];
}
roffs += layout.xSize[ifit];
}
chi2DataVec_.resize(layout.totalSize);
chi2ModVec_.resize(layout.totalSize);
chi2Vec_.resize(layout.totalSize);
fitVar_ = fitVar_.cwiseProduct(makeCorrFilter());
fitVarInv_ = fitVar_.pInverse(getSvdTolerance());
scheduleFitVarMatInit(false);
}
}
// buffer list of x vectors ////////////////////////////////////////////////////
void XYStatData::updateXMap(void) const
{
if (initXMap_)
{
XYStatData * modThis = const_cast<XYStatData *>(this);
modThis->xMap_.clear();
modThis->xMap_.resize(getMaxDataIndex());
for (auto k: getDataIndexSet())
{
modThis->xMap_[k] = DVec(getNXDim());
for (Index i = 0; i < getNXDim(); ++i)
{
modThis->xMap_[k](i) = xData_[i](dataCoord(k)[i]);
}
}
modThis->initXMap_ = false;
}
}
// buffer chi^2 vectors ////////////////////////////////////////////////////////
void XYStatData::updateChi2DataVec(void)
{
if (initChi2DataVec_)
{
Index a = 0, j, k, i, r;
updateLayout();
for (Index jfit = 0; jfit < layout.nYFitDim; ++jfit)
for (Index sfit = 0; sfit < layout.ySize[jfit]; ++sfit)
{
j = layout.yDim[jfit];
k = layout.data[jfit][sfit];
chi2DataVec_(a) = yData_[j][k];
a++;
}
for (Index ifit = 0; ifit < layout.nXFitDim; ++ifit)
for (Index rfit = 0; rfit < layout.xSize[ifit]; ++rfit)
{
i = layout.xDim[ifit];
r = layout.x[ifit][rfit];
chi2DataVec_(a) = xData_[i](r);
a++;
}
initChi2DataVec_ = false;
}
}
// WARNING: updateChi2ModVec is heavily called by fit
void XYStatData::updateChi2ModVec(const DVec p,
const vector<const DoubleModel *> &v,
const Index nPar, const Index nXDim)
{
updateLayout();
updateXMap();
Index a = 0, j, k, ind;
auto &par = p.segment(0, nPar), &xsi = p.segment(nPar, layout.totalXSize);
for (Index jfit = 0; jfit < layout.nYFitDim; ++jfit)
{
j = layout.yDim[jfit];
for (Index sfit = 0; sfit < layout.ySize[jfit]; ++sfit)
{
k = layout.data[jfit][sfit];
for (Index i = 0; i < nXDim; ++i)
{
ind = layout.xIndFromData[k][i] - layout.totalYSize;
xBuf_(i) = (ind >= 0) ? xsi(ind) : xMap_[k](i);
}
chi2ModVec_(a) = (*v[j])(xBuf_.data(), par.data());
a++;
}
}
chi2ModVec_.segment(a, layout.totalXSize) = xsi;
}