ANASEN_analysis/GainMatchQQQ.C
Vignesh Sitaraman 97880940be modified: .gitignore
modified:   .vscode/settings.json
	modified:   Analyzer.C
	modified:   Calibration.C
	modified:   GainMatchQQQ.C
	modified:   QQQ_Calcheck.C
2025-12-16 15:41:21 -05:00

309 lines
10 KiB
C

#define GainMatchQQQ_cxx
#include "GainMatchQQQ.h"
#include <TH2.h>
#include <TF1.h>
#include <TStyle.h>
#include <TCanvas.h>
#include <TMath.h>
#include <TCutG.h>
#include <fstream>
#include <utility>
#include <algorithm>
#include <cmath>
#include <numeric>
#include "Armory/HistPlotter.h"
#include "TVector3.h"
#include "TGraphErrors.h"
#include "TF1.h"
#include <cmath>
TH2F *hQQQFVB;
HistPlotter *plotter;
int padID = 0;
TCutG *cut;
std::map<std::tuple<int, int, int>, std::vector<std::pair<double, double>>> dataPoints;
void GainMatchQQQ::Begin(TTree * /*tree*/)
{
plotter = new HistPlotter("GainQQQ.root", "TFILE");
TString option = GetOption();
hQQQFVB = new TH2F("hQQQFVB", "QQQ Front vs Back; Front E; Back E", 800, 0, 16000, 800, 0, 16000);
// Load the TCutG object
TFile *cutFile = TFile::Open("qqqcorr.root");
if (!cutFile || cutFile->IsZombie())
{
std::cerr << "Error: Could not open qqqcorr.root" << std::endl;
return;
}
cut = dynamic_cast<TCutG *>(cutFile->Get("qqqcorr"));
if (!cut)
{
std::cerr << "Error: Could not find TCutG named 'qqqcorr' in qqqcorr.root" << std::endl;
return;
}
cut->SetName("qqqcorr"); // Ensure the cut has the correct name
}
Bool_t GainMatchQQQ::Process(Long64_t entry)
{
int ringMults[16] = {0};
int wedgeMults[16] = {0};
std::vector<std::tuple<int, int, int, double, double>> events;
b_qqqMulti->GetEntry(entry);
b_qqqID->GetEntry(entry);
b_qqqCh->GetEntry(entry);
b_qqqE->GetEntry(entry);
b_qqqT->GetEntry(entry);
qqq.CalIndex();
for (int i = 0; i < qqq.multi; i++)
{
for (int j = i + 1; j < qqq.multi; j++)
{
if (qqq.id[i] == qqq.id[j])
{
int chWedge = -1;
int chRing = -1;
float eWedge = 0.0;
float eRing = 0.0;
if (qqq.ch[i] < 16 && qqq.ch[j] >= 16)
{
chWedge = qqq.ch[i];
eWedge = qqq.e[i];
chRing = qqq.ch[j] - 16;
eRing = qqq.e[j];
}
else if (qqq.ch[j] < 16 && qqq.ch[i] >= 16)
{
chWedge = qqq.ch[j];
eWedge = qqq.e[j];
chRing = qqq.ch[i] - 16;
eRing = qqq.e[i];
}
else
continue;
ringMults[chRing]++;
wedgeMults[chWedge]++;
hQQQFVB->Fill(eWedge, eRing);
events.emplace_back(qqq.id[i], chRing, chWedge, eRing, eWedge);
plotter->Fill2D(Form("hRaw_qqq%d_ring%d_wedge%d", qqq.id[i], chRing, chWedge), 800, 0, 3000, 800, 0, 3000, eWedge, eRing, "hRawQQQ");
// double ratio = (eWedge > 0.0) ? (eRing / eWedge) : 0.0;
// double maxslope = 1.5;
// bool validPoint = false;
// if (ratio < maxslope && ratio > 1. / maxslope)
// {
// // Accumulate data for gain matching
// dataPoints[{qqq.id[i], chRing, chWedge}].emplace_back(eWedge, eRing);
// plotter->Fill2D("hAll_in", 4000, 0, 16000, 4000, 0, 16000, eWedge, eRing);
// validPoint = true;
// }
// if (!validPoint)
// {
// plotter->Fill2D("hAll_out", 4000, 0, 16000, 4000, 0, 16000, eWedge, eRing);
// }
}
}
}
for (auto tuple : events)
{
auto [id, chr, chw, er, ew] = tuple;
if (ringMults[chr] > 1 || wedgeMults[chw] > 1)
continue; // ignore multiplicity > 1 events
double ratio = (ew > 0.0) ? (er / ew) : 0.0;
double maxslope = 1.5;
bool validPoint = false;
if (ratio < maxslope && ratio > 1. / maxslope)
{
// Accumulate data for gain matching
dataPoints[{id, chr, chw}].emplace_back(ew, er);
plotter->Fill2D("hAll_in", 4000, 0, 16000, 4000, 0, 16000, ew, er);
validPoint = true;
}
if (!validPoint)
{
plotter->Fill2D("hAll_out", 4000, 0, 16000, 4000, 0, 16000, ew, er);
}
}
return kTRUE;
}
void GainMatchQQQ::Terminate()
{
const int MAX_DET = 4;
const int MAX_RING = 16;
const int MAX_WEDGE = 16;
// We store gains locally just for the "corrected" plot,
// but the file will output Slopes for the global minimizer.
double gainW[MAX_DET][MAX_RING][MAX_WEDGE] = {{{0}}};
double gainR[MAX_DET][MAX_RING][MAX_WEDGE] = {{{0}}};
bool gainValid[MAX_DET][MAX_RING][MAX_WEDGE] = {{{false}}};
// Output file for the Minimizer
std::ofstream outFile("qqq_GainMatch.txt");
// Benchmark/Debug file
std::ofstream benchFile("benchmark_diff.txt");
benchFile << "ID Wedge Ring Chi2NDF Slope SlopeErr" << std::endl;
if (!outFile.is_open()) { std::cerr << "Error opening output file!" << std::endl; return; }
const int MIN_POINTS = 50;
const int MAX_ITER = 3; // Outlier rejection passes
const double CLIP_SIGMA = 2.5; // Sigma threshold for outliers
for (const auto &kv : dataPoints)
{
auto key = kv.first;
auto [id, ring, wedge] = key;
const auto &pts = kv.second;
if (pts.size() < (size_t)MIN_POINTS) continue;
std::vector<std::pair<double, double>> current_pts = pts;
double finalSlope = 0.0;
double finalSlopeErr = 0.0;
bool converged = false;
// --- Iterative Fitting ---
for (int iter = 0; iter < MAX_ITER; ++iter)
{
if (current_pts.size() < (size_t)MIN_POINTS) break;
std::vector<double> x, y, ex, ey;
for (const auto &p : current_pts)
{
x.push_back(p.first); // Wedge E
y.push_back(p.second); // Ring E
ex.push_back(std::sqrt(std::abs(p.first))); // Error in X (Poisson)
ey.push_back(std::sqrt(std::abs(p.second))); // Error in Y (Poisson)
// Sanity check to avoid 0 error
if(ex.back() < 1.0) ex.back() = 1.0;
if(ey.back() < 1.0) ey.back() = 1.0;
}
// 2. Create Graph
TGraphErrors *gr = new TGraphErrors(current_pts.size(), x.data(), y.data(), ex.data(), ey.data());
// 3. Fit Linear Function through Origin
TF1 *f1= new TF1("calibFit", "[0]*x", 0, 16000);
f1->SetParameter(0, 1.0);
// "Q"=Quiet, "N"=NoDraw, "S"=ResultPtr
// We do NOT use "W" (Ignore weights), we want to use the errors we set.
int fitStatus = gr->Fit(f1, "QNS");
if (fitStatus != 0) {
delete gr; delete f1;
break;
}
finalSlope = f1->GetParameter(0);
double chi2 = f1->GetChisquare();
double ndf = f1->GetNDF();
// Get the statistical error on the slope
double rawErr = f1->GetParError(0);
// SCALING ERROR:
// If Chi2/NDF > 1, the data scatters more than Poisson stats predict.
// // We inflate the error by sqrt(Chi2/NDF) to be conservative for the Minimizer.
// double redChi2 = (ndf > 0) ? (chi2 / ndf) : 1.0;
// double inflation = (redChi2 > 1.0) ? std::sqrt(redChi2) : 1.0;
// finalSlopeErr = rawErr * inflation;
// 4. Outlier Rejection
if (iter == MAX_ITER - 1) {
converged = true;
delete gr; delete f1;
break;
}
// Calculate Residuals
std::vector<double> residuals;
double sumSqResid = 0.0;
for(size_t k=0; k<current_pts.size(); ++k) {
double val = f1->Eval(current_pts[k].first);
double res = current_pts[k].second - val;
residuals.push_back(res);
sumSqResid += res*res;
}
// double sigma = std::sqrt(sumSqResid / current_pts.size());
// // Filter
// std::vector<std::pair<double, double>> next_pts;
// for(size_t k=0; k<current_pts.size(); ++k) {
// if(std::abs(residuals[k]) < CLIP_SIGMA * sigma) {
// next_pts.push_back(current_pts[k]);
// }
// }
// if (next_pts.size() == current_pts.size()) {
// converged = true;
// delete gr; delete f1;
// break;
// }
// current_pts = next_pts;
// delete gr; delete f1;
}
if (!converged || finalSlope <= 0) continue;
// --- Store/Output ---
// 1. Save locally for the verification plot (hAll)
// Approximate local gain for plotting purposes only
double gW_local = std::sqrt(finalSlope);
double gR_local = 1.0 / gW_local;
gainW[id][ring][wedge] = gW_local;
gainR[id][ring][wedge] = gR_local;
gainValid[id][ring][wedge] = true;
// 2. Write to File for Minimizer
// Format: ID Wedge Ring Slope Error
outFile << id << " " << wedge << " " << ring << " " << finalSlope << " " << finalSlopeErr << std::endl;
// 3. Benchmark Info
benchFile << id << " " << wedge << " " << ring << " "
<< finalSlope << " " << finalSlopeErr << std::endl;
}
outFile.close();
benchFile.close();
std::cout << "Gain matching with Errors complete." << std::endl;
// Plotting the corrected data (Visual check using local approx gains)
for (auto &kv : dataPoints)
{
int id, ring, wedge;
std::tie(id, ring, wedge) = kv.first;
if (!gainValid[id][ring][wedge]) continue;
auto &pts = kv.second;
for (auto &pr : pts)
{
double corrWedge = pr.first * gainW[id][ring][wedge];
double corrRing = pr.second * gainR[id][ring][wedge];
plotter->Fill2D("hAll", 4000, 0, 16000, 4000, 0, 16000, corrWedge, corrRing);
}
}
plotter->FlushToDisk();
}