Files
aare/include/aare/Interpolator.hpp
AliceMazzoleni99 b77a576f72
All checks were successful
Build on RHEL8 / build (push) Successful in 2m13s
Build on RHEL9 / build (push) Successful in 2m37s
Run tests using data on local RHEL8 / build (push) Successful in 3m12s
Dev/automate tests using data (#267)
- automatically run python tests 
- automatically run test using data files on local runner from gitea
- fixed some of the workflows

---------

Co-authored-by: Erik Fröjdh <erik.frojdh@psi.ch>
2026-01-20 17:20:48 +01:00

265 lines
9.4 KiB
C++

// SPDX-License-Identifier: MPL-2.0
#pragma once
#include "aare/CalculateEta.hpp"
#include "aare/Cluster.hpp"
#include "aare/ClusterFile.hpp" //Cluster_3x3
#include "aare/ClusterVector.hpp"
#include "aare/NDArray.hpp"
#include "aare/NDView.hpp"
#include "aare/algorithm.hpp"
namespace aare {
struct Photon {
double x;
double y;
double energy;
};
struct Coordinate2D {
double x{};
double y{};
};
class Interpolator {
/**
* @brief
* marginal CDF of eta_x (if rosenblatt applied), conditional
* CDF of eta_x conditioned on eta_y
* value at (i, j, e): F(eta_x[i] |
*eta_y[j], energy[e])
*/
NDArray<double, 3> m_ietax;
/**
* @brief
* conditional CDF of eta_y conditioned on eta_x
* value at (i,j,e): F(eta_y[j] | eta_x[i], energy[e])
*/
NDArray<double, 3> m_ietay;
NDArray<double, 1> m_etabinsx;
NDArray<double, 1> m_etabinsy;
NDArray<double, 1> m_energy_bins;
public:
/**
* @brief Constructor for the Interpolator class
* @param etacube joint distribution of etaX, etaY and photon energy (note
* first dimension is etaX, second etaY, third photon energy)
* @param xbins bin edges for etaX
* @param ybins bin edges for etaY
* @param ebins bin edges for photon energy
*/
Interpolator(NDView<double, 3> etacube, NDView<double, 1> xbins,
NDView<double, 1> ybins, NDView<double, 1> ebins);
/**
* @brief Constructor for the Interpolator class
* @param xbins bin edges for etaX
* @param ybins bin edges for etaY
* @param ebins bin edges for photon energy
*/
Interpolator(NDView<double, 1> xbins, NDView<double, 1> ybins,
NDView<double, 1> ebins);
/**
* @brief transforms the joint eta distribution of etaX and etaY to the two
* independant uniform distributions based on the Roseblatt transform for
* each energy level
* @param etacube joint distribution of etaX, etaY and photon energy (first
* dimension is etaX, second etaY, third photon energy)
*/
void rosenblatttransform(NDView<double, 3> etacube);
NDArray<double, 3> get_ietax() { return m_ietax; }
NDArray<double, 3> get_ietay() { return m_ietay; }
/**
* @brief interpolates the cluster centers for all clusters to a better
* precision
* @tparam ClusterType Type of Clusters to interpolate
* @tparam Etafunction Function object that calculates desired eta default:
* calculate_eta2
* @return interpolated photons (photon positions are given as double but
* following row column format e.g. x=0, y=0 means top row and first column
* of frame) (An interpolated photon position of (1.5, 2.5) corresponds to
* an estimated photon hit at the pixel center of pixel (1,2))
*/
template <auto EtaFunction = calculate_eta2, typename ClusterType,
typename Eanble = std::enable_if_t<is_cluster_v<ClusterType>>>
std::vector<Photon>
interpolate(const ClusterVector<ClusterType> &clusters) const;
/**
* @brief transforms the eta values to uniform coordinates based on the CDF
* ieta_x and ieta_y
* @tparam eta Eta to transform
* @return uniform coordinates {x,y}
*/
template <typename T>
Coordinate2D transform_eta_values(const Eta2<T> &eta) const;
private:
/**
* @brief bilinear interpolation of the transformed eta values
* @param ix index of etaX bin
* @param iy index of etaY bin
* @param ie index of energy bin
* @return pair of interpolated transformed eta values (ietax, ietay)
*/
template <typename T>
std::pair<double, double>
bilinear_interpolation(const size_t ix, const size_t iy, const size_t ie,
const Eta2<T> &eta) const;
};
template <typename T>
std::pair<double, double>
Interpolator::bilinear_interpolation(const size_t ix, const size_t iy,
const size_t ie,
const Eta2<T> &eta) const {
auto next_index_y = static_cast<ssize_t>(iy + 1) >= m_ietax.shape(1)
? m_ietax.shape(1) - 1
: iy + 1;
auto next_index_x = static_cast<ssize_t>(ix + 1) >= m_ietax.shape(0)
? m_ietax.shape(0) - 1
: ix + 1;
// bilinear interpolation
double ietax_interp_left = linear_interpolation(
{m_etabinsy(iy), m_etabinsy(iy + 1)},
{m_ietax(ix, iy, ie), m_ietax(ix, next_index_y, ie)}, eta.y);
double ietax_interp_right =
linear_interpolation({m_etabinsy(iy), m_etabinsy(iy + 1)},
{m_ietax(next_index_x, iy, ie),
m_ietax(next_index_x, next_index_y, ie)},
eta.y);
// transformed photon position x between [0,1]
double ietax_interpolated =
linear_interpolation({m_etabinsx(ix), m_etabinsx(ix + 1)},
{ietax_interp_left, ietax_interp_right}, eta.x);
double ietay_interp_left = linear_interpolation(
{m_etabinsx(ix), m_etabinsx(ix + 1)},
{m_ietay(ix, iy, ie), m_ietay(next_index_x, iy, ie)}, eta.x);
double ietay_interp_right =
linear_interpolation({m_etabinsx(ix), m_etabinsx(ix + 1)},
{m_ietay(ix, next_index_y, ie),
m_ietay(next_index_x, next_index_y, ie)},
eta.x);
// transformed photon position y between [0,1]
double ietay_interpolated =
linear_interpolation({m_etabinsy(iy), m_etabinsy(iy + 1)},
{ietay_interp_left, ietay_interp_right}, eta.y);
return {ietax_interpolated, ietay_interpolated};
}
template <typename T>
Coordinate2D Interpolator::transform_eta_values(const Eta2<T> &eta) const {
// Finding the index of the last element that is smaller
// should work fine as long as we have many bins
auto ie = last_smaller(m_energy_bins, static_cast<double>(eta.sum));
auto ix = last_smaller(m_etabinsx, eta.x);
auto iy = last_smaller(m_etabinsy, eta.y);
if (static_cast<ssize_t>(ix) >= m_etabinsx.size() - 1 ||
static_cast<ssize_t>(iy) >= m_etabinsy.size() - 1 ||
static_cast<ssize_t>(ie) >= m_energy_bins.size() - 1)
throw std::runtime_error(
fmt::format("Eta values out of bounds of eta distribution: eta.x = "
"{:.4f}, eta.y = {:.4f}, energy = {:.4f}",
eta.x, eta.y, eta.sum));
// TODO: bilinear interpolation only works if all bins have a size > 1 -
// otherwise bilinear interpolation with zero values which skew the
// results
// TODO: maybe trim the bins at the edges with zero values beforehand
// auto [ietax_interpolated, ietay_interpolated] =
// bilinear_interpolation(ix, iy, ie, eta);
return Coordinate2D{m_ietax(ix, iy, ie), m_ietay(ix, iy, ie)};
}
template <auto EtaFunction, typename ClusterType, typename Enable>
std::vector<Photon>
Interpolator::interpolate(const ClusterVector<ClusterType> &clusters) const {
std::vector<Photon> photons;
photons.reserve(clusters.size());
size_t cluster_index{};
for (const ClusterType &cluster : clusters) {
auto eta = EtaFunction(cluster);
Photon photon;
photon.x = cluster.x;
photon.y = cluster.y;
photon.energy = static_cast<decltype(photon.energy)>(eta.sum);
try {
// check if eta values are within bounds
transform_eta_values(eta);
} catch (const std::runtime_error &e) {
throw std::runtime_error(
fmt::format("{} for cluster: {}", e.what(), cluster_index));
}
auto uniform_coordinates = transform_eta_values(eta);
if (EtaFunction == &calculate_eta2<typename ClusterType::value_type,
ClusterType::cluster_size_x,
ClusterType::cluster_size_y,
typename ClusterType::coord_type> ||
EtaFunction ==
&calculate_full_eta2<typename ClusterType::value_type,
ClusterType::cluster_size_x,
ClusterType::cluster_size_y,
typename ClusterType::coord_type>) {
double dX{}, dY{};
// TODO: could also chaneg the sign of the eta calculation
switch (eta.c) {
case corner::cTopLeft:
dX = -1.0;
dY = -1.0;
break;
case corner::cTopRight:;
dX = 0.0;
dY = -1.0;
break;
case corner::cBottomLeft:
dX = -1.0;
dY = 0.0;
break;
case corner::cBottomRight:
dX = 0.0;
dY = 0.0;
break;
}
photon.x = photon.x + 0.5 + uniform_coordinates.x +
dX; // use pixel center + 0.5
photon.y =
photon.y + 0.5 + uniform_coordinates.y +
dY; // eta2 calculates the ratio between bottom and sum of
// bottom and top shift by 1 add eta value correctly
} else {
photon.x += uniform_coordinates.x;
photon.y += uniform_coordinates.y;
}
++cluster_index;
photons.push_back(photon);
}
return photons;
}
} // namespace aare