multipers 2.4.0b1__cp312-cp312-macosx_11_0_arm64.whl

multipers/multiparameter_module_approximation/approximation.h

@@ -0,0 +1,2541 @@
+
+/*    This file is part of the Gudhi Library - https://gudhi.inria.fr/ - which is released under MIT.
+ *    See file LICENSE or go to https://gudhi.inria.fr/licensing/ for full license details.
+ *    Author(s):       David Loiseaux
+ *
+ *    Copyright (C) 2021 Inria
+ *
+ *    Modification(s):
+ *      - 2022/03 Hannah Schreiber: Integration of the new Vineyard_persistence class, renaming and cleanup.
+ *      - 2022/05 Hannah Schreiber: Addition of Summand class and Module class.
+ */
+
+/**
+ * @file approximation.h
+ * @author David Loiseaux, Hannah Schreiber
+ * @brief Contains the functions related to the approximation of n-modules.
+ */
+
+#ifndef APPROXIMATION_H_INCLUDED
+#define APPROXIMATION_H_INCLUDED
+
+#include <algorithm>
+#include <cmath>
+#include <cstddef>
+#include <cstdint>
+#include <iostream>
+#include <iterator>
+#include <limits>
+#include <oneapi/tbb/parallel_for.h>
+#include <memory>
+#include <string>
+#include <type_traits>
+#include <utility>
+#include <vector>
+#include <ranges>
+
+#include "../gudhi/Persistence_slices_interface.h"
+#include "../gudhi/gudhi/Dynamic_multi_parameter_filtration.h"
+#include "../gudhi/gudhi/Multi_persistence/Line.h"
+#include "../gudhi/gudhi/Multi_persistence/Box.h"
+#include "../tensor/tensor.h"
+#include "gudhi/Multi_filtration/multi_filtration_utils.h"
+#include "utilities.h"
+#include "debug.h"
+
+namespace Gudhi {
+namespace multiparameter {
+namespace mma {
+
+using Debug::Timer;
+using Gudhi::multi_persistence::Box;
+using Gudhi::multi_persistence::Line;
+template <typename T>
+class Module;
+template <typename T>
+class Summand;
+
+void threshold_filters_list(std::vector<value_type> &filtersList, const Box<value_type> &box);
+
+template <typename value_type>
+class Module {
+ public:
+  using filtration_type = typename Summand<value_type>::filtration_type;
+  using module_type = std::vector<Summand<value_type>>;
+  using image_type = std::vector<std::vector<value_type>>;
+  using get_2dpixel_value_function_type = std::function<value_type(const typename module_type::const_iterator,
+                                                                   const typename module_type::const_iterator,
+                                                                   value_type,
+                                                                   value_type)>;
+  using get_pixel_value_function_type = std::function<value_type(const typename module_type::const_iterator,
+                                                                 const typename module_type::const_iterator,
+                                                                 std::vector<value_type> &)>;
+
+  Module();
+  Module(Box<value_type> &box);
+
+  void resize(unsigned int size, int number_of_parameters);
+  Summand<value_type> &at(unsigned int index);
+  Summand<value_type> &operator[](size_t index);
+
+  const Summand<value_type> &operator[](const size_t index) const;
+  void add_barcode(const Line<value_type> &line,
+                   const std::vector<std::vector<std::array<value_type, 2>>> &barcode,
+                   bool threshold);
+  typename module_type::iterator begin();
+  typename module_type::iterator end();
+  typename module_type::const_iterator begin() const;
+  typename module_type::const_iterator end() const;
+
+  void clean();
+  void fill(const value_type precision);
+
+  std::vector<image_type> get_vectorization(const value_type delta,
+                                            const value_type p,
+                                            const bool normalize,
+                                            const Gudhi::multi_persistence::Box<value_type> &box,
+                                            unsigned int horizontalResolution,
+                                            unsigned int verticalResolution);
+  std::vector<image_type> get_vectorization(unsigned int horizontalResolution,
+                                            unsigned int verticalResolution,
+                                            get_2dpixel_value_function_type get_pixel_value) const;
+  image_type get_vectorization_in_dimension(const dimension_type dimension,
+                                            const value_type delta,
+                                            const value_type p,
+                                            const bool normalize,
+                                            const Gudhi::multi_persistence::Box<value_type> &box,
+                                            unsigned int horizontalResolution,
+                                            unsigned int verticalResolution);
+  image_type get_vectorization_in_dimension(const dimension_type dimension,
+                                            unsigned int horizontalResolution,
+                                            unsigned int verticalResolution,
+                                            get_2dpixel_value_function_type get_pixel_value) const;
+  std::vector<value_type> get_landscape_values(const std::vector<value_type> &x, const dimension_type dimension) const;
+
+  template <typename value_type2>
+  friend Module<value_type2> direct_sum(const Module<value_type2> &a, const Module<value_type2> &b);
+
+  void add_modules_from_ptr(const std::vector<intptr_t> &modules) {
+    size_t total_size = this->size();
+    for (const auto &m : modules) {
+      total_size += reinterpret_cast<Module<value_type> *>(m)->size();
+    }
+    this->module_.resize(total_size);
+    size_t index = this->size();
+    for (const auto &m : modules) {
+      std::copy(reinterpret_cast<Module<value_type> *>(m)->module_.begin(),
+                reinterpret_cast<Module<value_type> *>(m)->module_.end(),
+                this->module_.begin() + index);
+      index += reinterpret_cast<Module<value_type> *>(m)->size();
+    }
+    for (const auto &m_ptr : modules) {
+      Module<value_type> *m = reinterpret_cast<Module<value_type> *>(m_ptr);
+      this->module_.insert(this->module_.end(), m->module_.begin(), m->module_.end());
+    }
+  };
+
+  image_type get_landscape(const dimension_type dimension,
+                           const unsigned int k,
+                           const Box<value_type> &box,
+                           const std::vector<unsigned int> &resolution) const;
+  std::vector<image_type> get_landscapes(const dimension_type dimension,
+                                         const std::vector<unsigned int> ks,
+                                         const Box<value_type> &box,
+                                         const std::vector<unsigned int> &resolution) const;
+  void add_summand(Summand<value_type> summand, int degree = -1);
+  Box<value_type> get_box() const;
+  void set_box(Box<value_type> box);
+  unsigned int size() const;
+  void infer_box(std::vector<filtration_type> &filters_list);
+  dimension_type get_dimension() const;
+  module_type get_summands_of_dimension(const int dimension) const;
+  std::vector<std::pair<std::vector<std::vector<value_type>>, std::vector<std::vector<value_type>>>>
+  get_corners_of_dimension(const int dimension) const;
+  MultiDiagram<multipers::tmp_interface::Filtration_value<value_type>, value_type>
+  get_barcode(const Line<value_type> &l, const dimension_type dimension = -1, const bool threshold = false) const;
+  std::vector<std::vector<std::pair<value_type, value_type>>> get_barcode2(const Line<value_type> &l,
+                                                                           const dimension_type dimension) const;
+  std::vector<std::vector<std::vector<std::pair<value_type, value_type>>>> get_barcodes2(
+      const std::vector<Line<value_type>> &lines,
+      const dimension_type dimension = -1) const;
+  MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> get_barcodes(
+      const std::vector<Line<value_type>> &lines,
+      const dimension_type dimension = -1,
+      const bool threshold = false) const;
+  MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> get_barcodes(
+      const std::vector<typename Line<value_type>::Point_t> &basepoints,
+      const dimension_type dimension = -1,
+      const bool threshold = false) const;
+  std::vector<int> euler_curve(const std::vector<multipers::tmp_interface::Filtration_value<value_type>> &points) const;
+
+  Module<value_type> permute_summands(const std::vector<int> &permutation) {
+    Module<value_type> out;
+    out.set_box(this->get_box());
+    out.resize(this->size(), this->box_.get_dimension());
+    for (size_t i = 0; i < permutation.size(); ++i) {
+      out[i] = this->module_[permutation[i]];
+    }
+    return out;
+  }
+
+  inline Box<value_type> get_bounds() const;
+  inline void rescale(const std::vector<value_type> &rescale_factors, int degree);
+  inline void translate(const std::vector<value_type> &translation, int degree);
+
+  std::vector<std::vector<value_type>> compute_pixels(const std::vector<std::vector<value_type>> &coordinates,
+                                                      const std::vector<int> &degrees,
+                                                      const Box<value_type> &box = {},
+                                                      const value_type delta = 0.1,
+                                                      const value_type p = 1,
+                                                      const bool normalize = true,
+                                                      const int n_jobs = 0);
+
+  std::vector<value_type> get_interleavings(const Box<value_type> &box);
+  using distance_to_idx_type = std::vector<std::vector<int>>;
+  distance_to_idx_type compute_distance_idx_to(const std::vector<value_type> &pt, bool full) const;
+  std::vector<value_type> compute_distance_to(const std::vector<value_type> &pt, bool negative) const;
+  std::vector<std::vector<value_type>> compute_distances_to(const std::vector<std::vector<value_type>> &pt,
+                                                            bool negative,
+                                                            int n_jobs) const;
+  template <typename dtype = value_type, typename indices_type = int32_t>
+  void inline compute_distances_to(dtype *data_ptr,
+                                   const std::vector<std::vector<value_type>> &pts,
+                                   bool negative,
+                                   int n_jobs) const;
+  using distances_to_idx_type = std::vector<distance_to_idx_type>;
+  distances_to_idx_type compute_distances_idx_to(const std::vector<std::vector<value_type>> &pt,
+                                                 bool full,
+                                                 int n_jobs) const;
+  std::vector<value_type> compute_pixels_of_degree(const typename module_type::iterator start,
+                                                   const typename module_type::iterator end,
+                                                   const value_type delta,
+                                                   const value_type p,
+                                                   const bool normalize,
+                                                   const Box<value_type> &box,
+                                                   const std::vector<std::vector<value_type>> &coordinates,
+                                                   const int n_jobs = 0);
+
+  using idx_dump_type =
+      std::vector<std::vector<std::pair<std::vector<std::vector<int>>, std::vector<std::vector<int>>>>>;
+  idx_dump_type to_idx(const std::vector<std::vector<value_type>> &grid) const;
+  Module<int64_t> grid_squeeze(const std::vector<std::vector<value_type>> &grid) const;
+
+  void snap_to_integers() {
+    for (auto &summand : module_) {
+      summand.snap_to_integers();
+    }
+  }
+
+  std::vector<std::vector<std::vector<int>>> to_flat_idx(const std::vector<std::vector<value_type>> &grid) const;
+
+  void evaluate_in_grid(const std::vector<std::vector<value_type>> &grid) {
+    // for (auto &summand : module_) {
+    // summand.evaluate_in_grid(grid);
+    // }
+    tbb::parallel_for(size_t(0), module_.size(), [&](size_t i) { module_[i].evaluate_in_grid(grid); });
+  }
+
+  std::vector<int> inline get_degree_splits() const;
+
+  inline friend bool operator==(const Module &a, const Module &b) {
+    if (a.get_dimension() != b.get_dimension()) return false;
+    if (a.box_ != b.box_) return false;
+    if (a.size() != b.size()) return false;
+    for (auto i : std::ranges::views::iota(0u, a.size())) {
+      if (a[i] != b[i]) return false;
+    }
+    return true;
+  }
+
+ private:
+  module_type module_;
+  Box<value_type> box_;
+  void _compute_2D_image(image_type &image,
+                         const typename module_type::iterator start,
+                         const typename module_type::iterator end,
+                         const value_type delta = 0.1,
+                         const value_type p = 1,
+                         const bool normalize = true,
+                         const Box<value_type> &box = Box<value_type>(),
+                         const unsigned int horizontalResolution = 100,
+                         const unsigned int verticalResolution = 100);
+  void _compute_2D_image(image_type &image,
+                         const typename module_type::const_iterator start,
+                         const typename module_type::const_iterator end,
+                         unsigned int horizontalResolution,
+                         unsigned int verticalResolution,
+                         get_2dpixel_value_function_type get_pixel_value) const;
+
+  value_type _get_pixel_value(const typename module_type::iterator start,
+                              const typename module_type::iterator end,
+                              const filtration_type x,
+                              const value_type delta,
+                              const value_type p,
+                              const bool normalize,
+                              const value_type moduleWeight) const;
+
+  void _add_bar_with_threshold(const Line<value_type> &line,
+                               const std::array<value_type, 2> &bar,
+                               const bool threshold_in,
+                               Summand<value_type> &summand);
+};
+
+template <typename value_type>
+class Summand {
+ public:
+  using births_type = Gudhi::multi_filtration::Dynamic_multi_parameter_filtration<value_type, false, false>;
+  using deaths_type = Gudhi::multi_filtration::Dynamic_multi_parameter_filtration<value_type, true, false>;
+  using filtration_type =
+      Gudhi::multi_filtration::Dynamic_multi_parameter_filtration<value_type, false, true>;  // same for death
+  using dimension_type = int;
+  Summand(int number_of_parameters = 1);
+  Summand(const births_type &birth_corners, const deaths_type &death_corners, dimension_type dimension);
+  Summand(const std::vector<value_type> &birth_corners,
+          const std::vector<value_type> &death_corners,
+          int num_parameters,
+          dimension_type dimension);
+
+  value_type get_interleaving() const;
+  value_type get_interleaving(const Box<value_type> &box);
+  value_type get_local_weight(const filtration_type &x, const value_type delta) const;
+
+  value_type distance_to_upper(const filtration_type &x, bool negative) const;
+  value_type distance_to_lower(const filtration_type &x, bool negative) const;
+  value_type distance_to(const filtration_type &x, bool negative) const;
+  std::tuple<int, int> distance_idx_to_upper(const filtration_type &x) const;
+  std::tuple<int, int> distance_idx_to_lower(const filtration_type &x) const;
+  std::vector<int> distance_idx_to(const filtration_type &x, bool full) const;
+  std::pair<multipers::tmp_interface::Filtration_value<value_type>,
+            multipers::tmp_interface::Filtration_value<value_type>>
+  get_bar(const Line<value_type> &line) const;
+  std::pair<value_type, value_type> get_bar2(const Line<value_type> &l) const;
+  void add_bar(value_type baseBirth,
+               value_type baseDeath,
+               const typename Line<value_type>::Point_t &basepoint,
+               filtration_type &birth,
+               filtration_type &death,
+               const bool threshold,
+               const Box<value_type> &box);
+  void add_bar(const filtration_type &birth, const filtration_type &death);
+  void add_bar(const typename Line<value_type>::Point_t &basepoint,
+               value_type birth,
+               value_type death,
+               const Box<value_type> &);
+
+  std::vector<filtration_type> _get_birth_list() const;
+  std::vector<filtration_type> _get_death_list() const;
+  std::vector<multipers::tmp_interface::Filtration_value<value_type>> get_birth_list() const;
+  std::vector<multipers::tmp_interface::Filtration_value<value_type>> get_death_list() const;
+  const births_type &get_upset() const;
+  const deaths_type &get_downset() const;
+  void clean();
+
+  void complete_birth(const value_type precision);
+  void complete_death(const value_type precision);
+
+  dimension_type get_dimension() const;
+  void set_dimension(dimension_type dimension);
+
+  value_type get_landscape_value(const std::vector<value_type> &x) const;
+
+  friend void swap(Summand &sum1, Summand &sum2) {
+    std::swap(sum1.birth_corners_, sum2.birth_corners_);
+    std::swap(sum1.death_corners_, sum2.death_corners_);
+    std::swap(sum1.distanceTo0_, sum2.distanceTo0_);
+    // 	std::swap(sum1.updateDistance_, sum2.updateDistance_);
+  };
+
+  friend bool operator==(const Summand &a, const Summand &b) {
+    return a.dimension_ == b.dimension_ && a.birth_corners_ == b.birth_corners_ && a.death_corners_ == b.death_corners_;
+  }
+
+  bool contains(const filtration_type &x) const;
+
+  Box<value_type> get_bounds() const {
+    if (birth_corners_.num_generators() == 0) return Box<value_type>();
+    auto dimension = birth_corners_.num_parameters();
+    typename Box<value_type>::Point_t m(dimension, std::numeric_limits<value_type>::infinity());
+    typename Box<value_type>::Point_t M(dimension, -std::numeric_limits<value_type>::infinity());
+    for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        m[p] = std::min(m[p], birth_corners_(g, p));
+      }
+    }
+    for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        auto corner_i = death_corners_(g, p);
+        if (corner_i != filtration_type::T_inf) M[p] = std::max(M[p], corner_i);
+      }
+    }
+    return Box(m, M);
+  }
+
+  void rescale(const std::vector<value_type> &rescale_factors) {
+    if (birth_corners_.num_generators() == 0) return;
+    auto dimension = birth_corners_.num_parameters();
+    for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        birth_corners_(g, p) *= rescale_factors[p];
+      }
+    }
+    for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        death_corners_(g, p) *= rescale_factors[p];
+      }
+    }
+  }
+
+  void translate(const std::vector<value_type> &translation) {
+    if (birth_corners_.num_generators() == 0) return;
+    auto dimension = birth_corners_.num_parameters();
+    for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        birth_corners_(g, p) += translation[p];
+      }
+    }
+    for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        death_corners_(g, p) += translation[p];
+      }
+    }
+  }
+
+  Summand<int64_t> grid_squeeze(const std::vector<std::vector<value_type>> &grid) const;
+
+  void evaluate_in_grid(const std::vector<std::vector<value_type>> &grid) {
+    if (birth_corners_.num_generators() == 0) return;
+    auto num_parameters = birth_corners_.num_parameters();
+    auto snap = [](value_type x) {
+      value_type a = std::floor(x);
+      value_type b = std::ceil(x);
+      size_t out;
+      if (x - a < b - x)
+        out = static_cast<size_t>(a);
+      else
+        out = static_cast<size_t>(b);
+      return out;
+    };
+    auto todo = [&](auto &corners) {
+      return [&](size_t g) {
+        // auto &x = corners[i];
+        // for (auto g = 0u; g < x.num_generators(); ++g) {
+        for (auto p = 0u; p < corners.num_parameters(); ++p) {
+          value_type snapped = snap(corners(g, p));
+          corners(g, p) = (snapped >= grid[p].size() ? filtration_type::inf(num_parameters)(0,p) : grid[p][snapped]);
+        }
+        // }
+      };
+    };
+
+    tbb::parallel_for(size_t(0), birth_corners_.size(), todo(birth_corners_));
+    if (death_corners_.num_generators() == 0) return;
+    tbb::parallel_for(size_t(0), death_corners_.size(), todo(death_corners_));
+  };
+
+  void snap_to_integers() {
+    if (birth_corners_.num_generators() == 0) return;
+    auto dimension = birth_corners_.num_parameters();
+    auto snap = [](value_type x) {
+      value_type a = std::floor(x);
+      value_type b = std::ceil(x);
+      if (x - a < b - x)
+        return a;
+      else
+        return b;
+    };
+    for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        auto &x = birth_corners_(g, p);
+        x = snap(x);
+      }
+    }
+    for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+      for (unsigned int p = 0; p < dimension; ++p) {
+        auto &x = death_corners_(g, p);
+        x = snap(x);
+      }
+    }
+  }
+
+ private:
+  using Generator = typename births_type::Generator;
+
+  births_type birth_corners_;
+  deaths_type death_corners_;
+  value_type distanceTo0_;
+  dimension_type dimension_;
+
+  void _compute_interleaving(const Box<value_type> &box);
+  void _add_birth(const filtration_type &birth);
+  void _add_death(const filtration_type &death);
+  value_type _rectangle_volume(const filtration_type &a, const filtration_type &b) const;
+  value_type _get_max_diagonal(const filtration_type &a, const filtration_type &b, const Box<value_type> &box) const;
+  value_type d_inf(const Generator &a, const Generator &b) const;
+  void _factorize_min(Generator &a, const Generator &b);
+  void _factorize_max(Generator &a, const Generator &b);
+  static void _clean(std::vector<filtration_type> &list, bool keep_inf = true);
+
+  static inline void _clean(births_type &list, bool keep_inf = true) { list.remove_empty_generators(keep_inf); }
+
+  static inline void _clean(deaths_type &list, bool keep_inf = true) { list.remove_empty_generators(keep_inf); }
+};
+
+inline void threshold_filters_list(std::vector<filtration_type> &filtersList, const Box<value_type> &box) {
+  return;
+  for (unsigned int i = 0; i < filtersList.size(); i++) {
+    for (unsigned int p = 0; p < filtersList[i].num_parameters(); ++p) {
+      value_type &value = filtersList[i](0, p);
+      value = std::min(std::max(value, box.get_lower_corner()[i]), box.get_upper_corner()[i]);
+    }
+  }
+}
+
+template <class Filtration_value, int axis = 0, bool sign = true>
+class LineIterator {
+ public:
+  using value_type = typename Filtration_value::value_type;
+  LineIterator(const typename Line<value_type>::Point_t &basepoint,
+               const typename Line<value_type>::Point_t &direction,
+               value_type precision,
+               int num_iterations)
+      : precision(precision), remaining_iterations(num_iterations), current_line(std::move(basepoint), direction) {};
+
+  LineIterator<Filtration_value, axis, sign> &operator++() {
+    //
+    auto &basepoint = current_line.base_point();
+    if (this->is_finished()) return *this;
+    // If we didn't reached the end, go to the next line
+    basepoint[axis] += sign ? precision : -precision;
+    --remaining_iterations;
+    return *this;
+  }
+
+  const Line<value_type> &operator*() const { return current_line; }
+
+  LineIterator<Filtration_value, axis, sign> &next(std::size_t i) {
+    auto &basepoint = current_line.base_point();
+    if (this->is_finished()) return *this;
+    // If we didn't reached the end, go to the next line
+    basepoint[i] += sign ? precision : -precision;
+    --remaining_iterations;
+    return *this;
+  }
+
+  bool is_finished() const { return remaining_iterations <= 0; }
+
+ private:
+  const value_type precision;
+  int remaining_iterations;
+  Line<value_type> current_line;
+};
+
+template <class Filtration_value, int axis_ = 0, bool sign = true, class Slicer>
+inline void __add_vineyard_trajectory_to_module(Module<typename Filtration_value::value_type> &module,
+                                                Slicer &&slicer,
+                                                LineIterator<Filtration_value, axis_, sign> &line_iterator,
+                                                const bool threshold,
+                                                int axis = 0) {
+  static_assert(std::is_same_v<typename Filtration_value::value_type, typename Slicer::Filtration_value::value_type>);
+  using value_type = typename Filtration_value::value_type;
+  // Line iterator should be on the biggest axis
+  constexpr const bool verbose = false;
+  constexpr const bool verbose2 = false;
+  while (!line_iterator.is_finished()) {
+    const Line<value_type> &new_line = (axis_ >= 0) ? *(++line_iterator) : *line_iterator.next(axis);
+    // if constexpr (axis_ >= 0) {
+    //   new_line = *(++line_iterator); // first line is always a persistence
+    // } else {
+    //   new_line = *line_iterator.next(axis);
+    // }
+    // copy, no need to add it
+    if constexpr (verbose) std::cout << "----------------------------------------------" << std::endl;
+    if constexpr (verbose) std::cout << "Line basepoint " << new_line.base_point() << std::endl;
+    slicer.push_to(new_line);
+
+    slicer.vineyard_update();
+    if constexpr (verbose2) std::cout << slicer << std::endl;
+    module.add_barcode(new_line, slicer.template get_flat_barcode<true>(), threshold);
+  };
+};
+
+template <class Filtration_value, class Slicer = multipers::tmp_interface::SimplicialVineMatrixTruc<>>
+void _rec_mma(Module<typename Filtration_value::value_type> &module,
+              typename Line<value_type>::Point_t &basepoint,
+              const std::vector<int> &grid_size,
+              int dim_to_iterate,
+              Slicer &&current_persistence,
+              const value_type precision,
+              bool threshold) {
+  if (dim_to_iterate <= 0) {
+    LineIterator<Filtration_value, 0> line_iterator(std::move(basepoint), precision, grid_size[0]);
+    __add_vineyard_trajectory_to_module<Filtration_value, 0, Slicer>(
+        module, std::move(current_persistence), line_iterator, threshold);
+    return;
+  }
+  Slicer pers_copy;
+  typename Line<value_type>::Point_t basepoint_copy;
+  for (int i = 0; i < grid_size[dim_to_iterate]; ++i) {
+    // TODO : multithread, but needs matrix to be thread safe + put mutex on
+    // module
+    pers_copy = current_persistence;
+    basepoint_copy = basepoint;
+    _rec_mma(module, basepoint_copy, grid_size, dim_to_iterate - 1, pers_copy, precision, threshold);
+    basepoint[dim_to_iterate] += precision;
+    // current_persistence.push_to(Line(basepoint));
+    // current_persistence.vineyard_update();
+  }
+}
+
+template <int axis, class Filtration_value, class Slicer>
+void _rec_mma2(Module<typename Filtration_value::value_type> &module,
+               typename Line<typename Filtration_value::value_type>::Point_t &&basepoint,
+               const Filtration_value &direction,
+               const std::vector<int> &grid_size,
+               const std::vector<bool> &signs,
+               int dim_to_iterate,
+               Slicer &&current_persistence,
+               const value_type precision,
+               bool threshold) {
+  static_assert(std::is_same_v<typename Filtration_value::value_type, typename Slicer::T>);
+
+  if (dim_to_iterate <= axis) {
+    if (signs[axis]) {
+      LineIterator<Filtration_value, axis, true> line_iterator(
+          std::move(basepoint), direction, precision, grid_size[axis]);
+      __add_vineyard_trajectory_to_module<Filtration_value, axis, true, Slicer>(
+          module, std::move(current_persistence), line_iterator, threshold);
+    } else {
+      LineIterator<Filtration_value, axis, false> line_iterator(
+          std::move(basepoint), direction, precision, grid_size[axis]);
+      __add_vineyard_trajectory_to_module<Filtration_value, axis, false, Slicer>(
+          module, std::move(current_persistence), line_iterator, threshold);
+    }
+
+    return;
+  }
+  if (grid_size[dim_to_iterate] == 0) {
+    // no need to copy basepoint, we just skip the dim here
+    _rec_mma2<axis, Filtration_value, Slicer>(module,
+                                              std::move(basepoint),
+                                              direction,
+                                              grid_size,
+                                              signs,
+                                              dim_to_iterate - 1,
+                                              std::move(current_persistence),
+                                              precision,
+                                              threshold);
+    return;
+  }
+  for (int i = 0; i < grid_size[dim_to_iterate]; ++i) {
+    // TODO : multithread, but needs matrix to be thread safe + put mutex on
+    // module
+    _rec_mma2<axis, Filtration_value, typename Slicer::Thread_safe>(
+        module,
+        typename Line<typename Filtration_value::value_type>::Point_t(basepoint),
+        direction,
+        grid_size,
+        signs,
+        dim_to_iterate - 1,
+        current_persistence.weak_copy(),
+        precision,
+        threshold);
+    basepoint[dim_to_iterate] += signs[dim_to_iterate] ? precision : -precision;
+    // current_persistence.push_to(Line(basepoint));
+    // current_persistence.vineyard_update();
+  }
+}
+
+template <class Slicer, typename value_type>
+Module<value_type> multiparameter_module_approximation(Slicer &slicer,
+                                                       const typename Line<value_type>::Point_t &direction,
+                                                       const value_type precision,
+                                                       Box<value_type> &box,
+                                                       const bool threshold,
+                                                       const bool complete,
+                                                       const bool verbose) {
+  static_assert(std::is_same_v<typename Slicer::Filtration_value::value_type,
+                               value_type>);  // Value type can be exposed to python interface.
+  if (verbose) std::cout << "Starting Module Approximation" << std::endl;
+  /* using Filtration_value = Slicer::Filtration_value; */
+
+  typename Box<value_type>::Point_t basepoint = box.get_lower_corner();
+  const std::size_t num_parameters = box.get_dimension();
+  std::vector<int> grid_size(num_parameters);
+  std::vector<bool> signs(num_parameters);
+  int signs_shifts = 0;
+  int arg_max_signs_shifts = -1;
+  for (std::size_t i = 0; i < num_parameters; i++) {
+    auto &a = box.get_lower_corner()[i];
+    auto &b = box.get_upper_corner()[i];
+    grid_size[i] = static_cast<int>(std::ceil((std::fabs(b - a) / precision))) + 1;
+    signs[i] = b > a;
+    if (b < a) {
+      std::swap(a, b);
+      int local_shift;
+      if (!direction.size())
+        local_shift = grid_size[i];
+      else {
+        local_shift = direction[i] > 0 ? static_cast<int>(std::ceil(grid_size[i] / direction[i])) : 0;
+      }
+      if (local_shift > signs_shifts) {
+        signs_shifts = std::max(signs_shifts, local_shift);
+        arg_max_signs_shifts = i;
+      }
+    }
+
+    // fix the box
+  }
+  if (signs_shifts > 0) {
+    for (std::size_t i = 0; i < num_parameters; i++)
+      grid_size[i] += signs_shifts;  // this may be too much for large num_parameters
+    grid_size[arg_max_signs_shifts] = 1;
+    if (verbose)
+      std::cout << "Had to flatten/shift coordinate " << arg_max_signs_shifts << " by " << signs_shifts << std::endl;
+  }
+  Module<value_type> out(box);
+  box.inflate(2 * precision);  // for infinte summands
+
+  if (verbose) std::cout << "Num parameters : " << num_parameters << std::endl;
+  if (verbose) std::cout << "Box : " << box << std::endl;
+  if (num_parameters < 1) return out;
+
+  // first line to compute
+  // TODO: change here
+  // for (auto i = 0u; i < basepoint.size() - 1; i++)
+  //   basepoint[i] -= box.get_upper_corner().back();
+  // basepoint.back() = 0;
+  Line<value_type> current_line(basepoint, direction);
+  if (verbose) std::cout << "First line basepoint " << basepoint << std::endl;
+
+  {
+    Timer timer("Initializing mma...\n", verbose);
+    // fills the first barcode
+    slicer.push_to(current_line);
+    slicer.initialize_persistence_computation(false);
+    auto barcode = slicer.template get_flat_barcode<true>();
+    auto num_bars = 0;
+    for (const auto &b : barcode) num_bars += b.size();
+    out.resize(num_bars, num_parameters);
+    std::size_t i = 0;
+    for (unsigned int dim = 0; dim < barcode.size(); ++dim) {
+      for ([[maybe_unused]] const auto &bar : barcode[dim]) {
+        out[i].set_dimension(dim);
+        ++i;
+      }
+    }
+    out.add_barcode(current_line, barcode, threshold);
+
+    if (verbose) std::cout << "Instantiated " << num_bars << " summands" << std::endl;
+  }
+  // TODO : change here
+  // std::vector<int> grid_size(num_parameters - 1);
+  // auto h = box.get_upper_corner().back() - box.get_lower_corner().back();
+  // for (int i = 0; i < num_parameters - 1; i++) {
+  //   auto a = box.get_lower_corner()[i];
+  //   auto b = box.get_upper_corner()[i];
+  //   grid_size[i] =
+  //       static_cast<unsigned int>(std::ceil((std::abs(b - a + h) /
+  //       precision)));
+  // }
+  // TODO : change here
+  if (verbose) {
+    std::cout << "Grid size ";
+    for (auto v : grid_size) std::cout << v << " ";
+    std::cout << " Signs ";
+    if (signs.empty()) {
+      std::cout << "[]";
+    } else {
+      std::cout << "[";
+      for (std::size_t i = 0; i < signs.size() - 1; i++) {
+        std::cout << signs[i] << ", ";
+      }
+      std::cout << signs.back() << "]";
+    }
+    std::cout << std::endl;
+    std::cout << "Max error " << precision << std::endl;
+  }
+
+  {
+    Timer timer("Computing mma...", verbose);
+    // actual computation. -1 as line grid is n-1 dim, -1 as we start from 0
+    // _rec_mma(out, basepoint, grid_size, num_parameters - 2, slicer,
+    // precision,
+    //          threshold);
+    // TODO : change here
+
+    for (std::size_t i = 1; i < num_parameters; i++) {
+      // the loop is on the faces of the lower box
+      // should be parallelizable, up to a mutex on out
+      if (direction.size() && direction[i] == 0.0) continue;  // skip faces with codim d_i=0
+      auto temp_grid_size = grid_size;
+      temp_grid_size[i] = 0;
+      if (verbose) {
+        std::cout << "Face " << i << "/" << num_parameters << " with grid size ";
+        for (auto v : temp_grid_size) std::cout << v << " ";
+        std::cout << std::endl;
+      }
+      // if (!direction.size() || direction[0] > 0)
+      _rec_mma2<0>(out,
+                   typename Line<value_type>::Point_t(basepoint),
+                   direction,
+                   temp_grid_size,
+                   signs,
+                   num_parameters - 1,
+                   slicer.weak_copy(),
+                   precision,
+                   threshold);
+    }
+    // last one, we can destroy basepoint & cie
+    if (!direction.size() || direction[0] > 0) {
+      grid_size[0] = 0;
+      if (verbose) {
+        std::cout << "Face " << num_parameters << "/" << num_parameters << " with grid size ";
+        for (auto v : grid_size) std::cout << v << " ";
+        std::cout << std::endl;
+      }
+      _rec_mma2<1>(out,
+                   std::move(basepoint),
+                   direction,
+                   grid_size,
+                   signs,
+                   num_parameters - 1,
+                   std::move(slicer),
+                   precision,
+                   threshold);
+    }
+  }
+
+  {  // for Timer
+    Timer timer("Cleaning output ... ", verbose);
+    out.clean();
+    if (complete) {
+      if (verbose) std::cout << "Completing output ...";
+      for (std::size_t i = 0; i < num_parameters; i++) out.fill(precision);
+    }
+  }  // Timer death
+  return out;
+};
+
+template <typename value_type>
+inline void Module<value_type>::add_barcode(const Line<value_type> &line,
+                                            const std::vector<std::vector<std::array<value_type, 2>>> &barcode,
+                                            const bool threshold_in) {
+#ifdef GUDHI_USE_TBB
+  std::vector<std::size_t> shifts(barcode.size(),0U);
+  for (std::size_t i = 1U; i < barcode.size(); i++) {
+    shifts[i] = shifts[i - 1] + barcode[i - 1].size();
+  }
+  tbb::parallel_for(size_t(0), barcode.size(), [&](size_t dim) {
+    tbb::parallel_for(size_t(0), barcode[dim].size(), [&](size_t j) {
+      _add_bar_with_threshold(line, barcode[dim][j], threshold_in, this->operator[](shifts[dim] + j));
+    });
+  });
+#else
+  auto count = 0U;
+  for (const auto &bar_dim : barcode) {
+    for (const auto &bar : bar_dim) _add_bar_with_threshold(line, bar, threshold_in, this->operator[](count++));
+  }
+#endif
+}
+
+template <typename value_type>
+inline void Module<value_type>::_add_bar_with_threshold(const Line<value_type> &line,
+                                                        const std::array<value_type, 2> &bar,
+                                                        const bool threshold_in,
+                                                        Summand<value_type> &summand) {
+  constexpr const bool verbose = false;
+  auto birth_filtration = bar[0], death_filtration = bar[1];
+
+  if (birth_filtration >= death_filtration) return;
+
+  if constexpr (verbose) {
+    std::cout << "--BAR (" << birth_filtration << ", " << death_filtration << ") at basepoint " << line.base_point()
+              << " direction " << line.direction() << std::endl;
+  }
+
+  filtration_type birth_container = line[birth_filtration];
+  if constexpr (verbose) std::cout << " B: " << birth_container << " B*d: " << birth_filtration * line.direction();
+  if (birth_container.is_minus_inf()) {
+    if (threshold_in) birth_container = box_.get_lower_corner();
+  } else {
+    bool allInf = true;
+    for (std::size_t i = 0U; i < birth_container.num_parameters(); i++) {
+      auto t = box_.get_lower_corner()[i];
+      if (birth_container(0, i) < t - 1e-10) birth_container(0, i) = threshold_in ? t : -filtration_type::T_inf;
+      if (birth_container(0, i) != -filtration_type::T_inf) allInf = false;
+    }
+    if (allInf) birth_container = filtration_type::minus_inf(birth_container.num_parameters());
+  }
+
+  filtration_type death_container = line[death_filtration];
+  if constexpr (verbose) std::cout << " D: " << death_container;
+  if (death_container.is_plus_inf()) {
+    if (threshold_in) death_container = box_.get_upper_corner();
+  } else {
+    bool allInf = true;
+    for (std::size_t i = 0U; i < death_container.num_parameters(); i++) {
+      auto t = box_.get_upper_corner()[i];
+      if (death_container(0, i) > t + 1e-10) death_container(0, i) = threshold_in ? t : filtration_type::T_inf;
+      if (death_container(0, i) != filtration_type::T_inf) allInf = false;
+    }
+    if (allInf) death_container = filtration_type::inf(death_container.num_parameters());
+  }
+
+  if constexpr (verbose) std::cout << " BT: " << birth_container << " DT: " << death_container << std::endl;
+  summand.add_bar(birth_container, death_container);
+}
+
+template <typename value_type>
+inline Module<value_type>::Module() {}
+
+template <typename value_type>
+inline Module<value_type>::Module(Box<value_type> &box) : box_(box) {}
+
+template <typename value_type>
+inline void Module<value_type>::resize(const unsigned int size, int number_of_parameters) {
+  module_.resize(size, Summand<value_type>(number_of_parameters));
+}
+
+template <typename value_type>
+inline Summand<value_type> &Module<value_type>::at(const unsigned int index) {
+  return module_.at(index);
+}
+
+template <typename value_type>
+inline Summand<value_type> &Module<value_type>::operator[](const size_t index) {
+  return this->module_[index];
+}
+
+template <typename value_type>
+inline const Summand<value_type> &Module<value_type>::operator[](const size_t index) const {
+  return this->module_[index];
+}
+
+template <typename value_type>
+inline typename Module<value_type>::module_type::iterator Module<value_type>::begin() {
+  return module_.begin();
+}
+
+template <typename value_type>
+inline typename Module<value_type>::module_type::const_iterator Module<value_type>::begin() const {
+  return module_.cbegin();
+}
+
+template <typename value_type>
+inline typename Module<value_type>::module_type::iterator Module<value_type>::end() {
+  return module_.end();
+}
+
+template <typename value_type>
+inline typename Module<value_type>::module_type::const_iterator Module<value_type>::end() const {
+  return module_.cend();
+}
+
+template <typename value_type>
+inline void Module<value_type>::add_summand(Summand<value_type> summand, int degree) {
+  if (degree >= 0) summand.set_dimension(degree);
+  module_.push_back(summand);
+}
+
+/**
+ * @brief Remove the empty summands of the output
+ *
+ * @param output p_output:...
+ * @param keep_order p_keep_order:... Defaults to false.
+ */
+
+template <typename value_type>
+inline void Module<value_type>::clean() {
+  module_type tmp;
+  for (size_t i = 0u; i < module_.size(); i++) {
+    module_[i].clean();
+  }
+  module_.erase(
+      std::remove_if(
+          module_.begin(), module_.end(), [](const Summand<value_type> &s) { return s.get_upset().is_plus_inf(); }),
+      module_.end());
+}
+
+template <typename value_type>
+inline void Module<value_type>::fill(const value_type precision) {
+  if (module_.empty()) return;
+
+  for (Summand<value_type> &sum : module_) {
+    sum.complete_birth(precision);
+    sum.complete_death(precision);
+  }
+}
+
+template <typename value_type>
+std::vector<value_type> inline Module<value_type>::get_interleavings(const Box<value_type> &box) {
+  std::vector<value_type> out(this->size());
+  for (auto i = 0u; i < out.size(); ++i) {
+    out[i] = module_[i].get_interleaving(box);
+  }
+  return out;
+}
+
+template <typename value_type>
+typename Module<value_type>::distance_to_idx_type inline Module<value_type>::compute_distance_idx_to(
+    const std::vector<value_type> &pt,
+    bool full) const {
+  typename Module<value_type>::distance_to_idx_type out(module_.size(), std::vector<int>(full ? 4 : 2));
+  for (auto i = 0u; i < module_.size(); ++i) {
+    out[i] = module_[i].distance_idx_to(pt, full);
+  }
+  return out;
+}
+
+template <typename value_type>
+std::vector<value_type> inline Module<value_type>::compute_distance_to(const std::vector<value_type> &pt,
+                                                                       bool negative) const {
+  std::vector<value_type> out(this->size());
+  for (auto i = 0u; i < this->size(); ++i) {
+    out[i] = module_[i].distance_to(pt, negative);
+  }
+  return out;
+}
+
+template <typename value_type>
+std::vector<std::vector<value_type>> inline Module<value_type>::compute_distances_to(
+    const std::vector<std::vector<value_type>> &pts,
+    bool negative,
+    int n_jobs) const {
+  std::vector<std::vector<value_type>> out(pts.size(), std::vector<value_type>(this->size()));
+  oneapi::tbb::task_arena arena(n_jobs);  // limits the number of threads
+  arena.execute([&] {
+    tbb::parallel_for(std::size_t(0u), pts.size(), [&](std::size_t i) {
+      tbb::parallel_for(std::size_t(0u), std::size_t(this->size()), [&](std::size_t j) {
+        out[i][j] = module_[j].distance_to(pts[i], negative);
+      });
+    });
+  });
+  return out;
+}
+
+template <typename value_type>
+template <typename dtype, typename indices_type>
+void inline Module<value_type>::compute_distances_to(dtype *data_ptr,
+                                                     const std::vector<std::vector<value_type>> &pts,
+                                                     bool negative,
+                                                     int n_jobs) const {
+  tensor::static_tensor_view<dtype, indices_type> container(
+      data_ptr, {static_cast<int>(pts.size()), static_cast<int>(this->size())});
+  oneapi::tbb::task_arena arena(n_jobs);  // limits the number of threads
+  arena.execute([&] {
+    tbb::parallel_for(std::size_t(0u), pts.size(), [&](std::size_t i) {
+      // tbb::parallel_for(std::size_t(0u), std::size_t(this->size()), [&](std::size_t j) {
+      dtype *current_ptr = &container[{static_cast<int>(i), 0}];
+      for (std::size_t j = 0u; j < this->size(); ++j) {
+        *(current_ptr + j) = module_[j].distance_to(pts[i], negative);
+      }
+    });
+    // });
+  });
+}
+
+template <typename value_type>
+typename Module<value_type>::distances_to_idx_type inline Module<value_type>::compute_distances_idx_to(
+    const std::vector<std::vector<value_type>> &pts,
+    bool full,
+    int n_jobs) const {
+  Module::distances_to_idx_type out(pts.size(),
+                                    Module::distance_to_idx_type(module_.size(), std::vector<int>(full ? 4 : 2)));
+
+  oneapi::tbb::task_arena arena(n_jobs);  // limits the number of threads
+  arena.execute([&] {
+    tbb::parallel_for(std::size_t(0u), pts.size(), [&](std::size_t i) {
+      tbb::parallel_for(std::size_t(0u), std::size_t(this->size()), [&](std::size_t j) {
+        out[i][j] = module_[j].distance_idx_to(pts[i], full);
+      });
+    });
+  });
+  return out;
+}
+
+template <typename value_type>
+inline std::vector<std::vector<value_type>> Module<value_type>::compute_pixels(
+    const std::vector<std::vector<value_type>> &coordinates,
+    const std::vector<int> &degrees,
+    const Box<value_type> &box,
+    const value_type delta,
+    const value_type p,
+    const bool normalize,
+    const int n_jobs) {
+  auto num_degrees = degrees.size();
+  auto num_pts = coordinates.size();
+  std::vector<std::vector<value_type>> out(num_degrees, std::vector<value_type>(num_pts));
+
+  typename module_type::iterator start;
+  typename module_type::iterator end = module_.begin();
+  for (auto degree_idx = 0u; degree_idx < num_degrees; degree_idx++) {
+    auto d = degrees[degree_idx];
+    {  // for Timer
+      Debug::Timer timer("Computing image of dimension " + std::to_string(d) + " ...", verbose);
+      start = end;
+      while (start != module_.end() && start->get_dimension() != d) start++;
+      if (start == module_.end()) break;
+      end = start;
+      while (end != module_.end() && end->get_dimension() == d) end++;
+      out[degree_idx] = compute_pixels_of_degree(start, end, delta, p, normalize, box, coordinates, n_jobs);
+    }  // Timer death
+  }
+  return out;
+}
+
+template <typename value_type>
+inline typename std::vector<typename Module<value_type>::image_type> Module<value_type>::get_vectorization(
+    const value_type delta,
+    const value_type p,
+    const bool normalize,
+    const Box<value_type> &box,
+    unsigned int horizontalResolution,
+    unsigned int verticalResolution) {
+  dimension_type maxDim = module_.back().get_dimension();
+  std::vector<Module::image_type> image_vector(maxDim + 1);
+  typename module_type::iterator start;
+  typename module_type::iterator end = module_.begin();
+  for (dimension_type d = 0; d <= maxDim; d++) {
+    {  // for Timer
+      Debug::Timer timer("Computing image of dimension " + std::to_string(d) + " ...", verbose);
+      start = end;
+      while (end != module_.end() && end->get_dimension() == d) end++;
+      _compute_2D_image(
+          image_vector.at(d), start, end, delta, p, normalize, box, horizontalResolution, verticalResolution);
+    }  // Timer death
+  }
+  return image_vector;
+}
+
+template <typename value_type>
+inline std::vector<typename Module<value_type>::image_type> Module<value_type>::get_vectorization(
+    unsigned int horizontalResolution,
+    unsigned int verticalResolution,
+    get_2dpixel_value_function_type get_pixel_value) const {
+  dimension_type maxDim = module_.back().get_dimension();
+  std::vector<Module::image_type> image_vector(maxDim + 1);
+  typename module_type::const_iterator start;
+  typename module_type::const_iterator end = module_.begin();
+  for (dimension_type d = 0; d <= maxDim; d++) {
+    {  // for Timer
+      Debug::Timer timer("Computing image of dimension " + std::to_string(d) + " ...", verbose);
+      start = end;
+      while (end != module_.end() && end->get_dimension() == d) end++;
+      _compute_2D_image(image_vector.at(d), start, end, horizontalResolution, verticalResolution, get_pixel_value);
+    }  // Timer death
+  }
+  return image_vector;
+}
+
+template <typename value_type>
+inline typename Module<value_type>::image_type Module<value_type>::get_vectorization_in_dimension(
+    const dimension_type dimension,
+    const value_type delta,
+    const value_type p,
+    const bool normalize,
+    const Box<value_type> &box,
+    unsigned int horizontalResolution,
+    unsigned int verticalResolution) {
+  Debug::Timer timer("Computing image of dimension " + std::to_string(dimension) + " ...", verbose);
+
+  Module::image_type image;
+  typename module_type::iterator start = module_.begin();
+  while (start != module_.end() && start->get_dimension() < dimension) start++;
+  typename module_type::iterator end = start;
+  while (end != module_.end() && end->get_dimension() == dimension) end++;
+  _compute_2D_image(image, start, end, delta, p, normalize, box, horizontalResolution, verticalResolution);
+
+  return image;
+}
+
+template <typename value_type>
+inline typename Module<value_type>::image_type Module<value_type>::get_vectorization_in_dimension(
+    const dimension_type dimension,
+    unsigned int horizontalResolution,
+    unsigned int verticalResolution,
+    get_2dpixel_value_function_type get_pixel_value) const {
+  Debug::Timer timer("Computing image of dimension " + std::to_string(dimension) + " ...", verbose);
+
+  typename Module::image_type image;
+  typename module_type::const_iterator start = module_.begin();
+  while (start != module_.end() && start->get_dimension() < dimension) start++;
+  typename module_type::const_iterator end = start;
+  while (end != module_.end() && end->get_dimension() == dimension) end++;
+  _compute_2D_image(image, start, end, horizontalResolution, verticalResolution, get_pixel_value);
+
+  return image;
+}
+
+template <typename value_type>
+std::vector<value_type> Module<value_type>::get_landscape_values(const std::vector<value_type> &x,
+                                                                 const dimension_type dimension) const {
+  std::vector<value_type> out;
+  out.reserve(this->size());
+  for (unsigned int i = 0; i < this->size(); i++) {
+    const Summand<value_type> &summand = this->module_[i];
+    if (summand.get_dimension() == dimension) out.push_back(summand.get_landscape_value(x));
+  }
+  std::sort(out.begin(), out.end(), [](const value_type x, const value_type y) { return x > y; });
+  return out;
+}
+
+template <typename value_type>
+typename Module<value_type>::image_type Module<value_type>::get_landscape(
+    const dimension_type dimension,
+    const unsigned int k,
+    const Box<value_type> &box,
+    const std::vector<unsigned int> &resolution) const {
+  // TODO extend in higher dimension (ie, change the image type to a template
+  // class)
+  Module::image_type image;
+  image.resize(resolution[0], std::vector<value_type>(resolution[1]));
+  value_type stepX = (box.get_upper_corner()[0] - box.get_lower_corner()[0]) / resolution[0];
+  value_type stepY = (box.get_upper_corner()[1] - box.get_lower_corner()[1]) / resolution[1];
+  tbb::parallel_for(0U, resolution[0], [&](unsigned int i) {
+    tbb::parallel_for(0U, resolution[1], [&](unsigned int j) {
+      auto landscape = this->get_landscape_values(
+          {box.get_lower_corner()[0] + stepX * i, box.get_lower_corner()[1] + stepY * j}, dimension);
+      image[i][j] = k < landscape.size() ? landscape[k] : 0;
+    });
+  });
+  return image;
+}
+
+template <typename value_type>
+std::vector<typename Module<value_type>::image_type> Module<value_type>::get_landscapes(
+    const dimension_type dimension,
+    const std::vector<unsigned int> ks,
+    const Box<value_type> &box,
+    const std::vector<unsigned int> &resolution) const {
+  std::vector<Module::image_type> images(ks.size());
+  for (auto &image : images) image.resize(resolution[0], std::vector<value_type>(resolution[1]));
+  value_type stepX = (box.get_upper_corner()[0] - box.get_lower_corner()[0]) / resolution[0];
+  value_type stepY = (box.get_upper_corner()[1] - box.get_lower_corner()[1]) / resolution[1];
+
+  tbb::parallel_for(0U, resolution[0], [&](unsigned int i) {
+    tbb::parallel_for(0U, resolution[1], [&](unsigned int j) {
+      std::vector<value_type> landscapes = this->get_landscape_values(
+          {box.get_lower_corner()[0] + stepX * i, box.get_lower_corner()[1] + stepY * j}, dimension);
+      for (const auto k : ks) {
+        images[k][i][j] = k < landscapes.size() ? landscapes[k] : 0;
+      }
+    });
+  });
+  return images;
+}
+
+template <typename value_type>
+inline Box<value_type> Module<value_type>::get_box() const {
+  return this->box_;
+}
+
+template <typename value_type>
+inline void Module<value_type>::set_box(Box<value_type> box) {
+  this->box_ = box;
+}
+
+template <typename value_type>
+inline unsigned int Module<value_type>::size() const {
+  return this->module_.size();
+}
+
+template <typename value_type>
+inline void Module<value_type>::infer_box(std::vector<filtration_type> &f) {
+  this->box_.infer_from_filters(f);
+}
+
+template <typename value_type>
+inline dimension_type Module<value_type>::get_dimension() const {
+  return this->module_.empty() ? -1 : this->module_.back().get_dimension();
+}
+
+template <typename value_type>
+inline std::vector<Summand<value_type>> Module<value_type>::get_summands_of_dimension(const int dimension) const {
+  std::vector<Summand<value_type>> list;
+  for (const Summand<value_type> &summand : this->module_) {
+    if (summand.get_dimension() == dimension) list.push_back(summand);
+  }
+  return list;
+}
+
+template <typename value_type>
+inline std::vector<std::pair<std::vector<std::vector<value_type>>, std::vector<std::vector<value_type>>>>
+Module<value_type>::get_corners_of_dimension(const int dimension) const {
+  std::vector<std::pair<std::vector<std::vector<value_type>>, std::vector<std::vector<value_type>>>> list;
+  for (const Summand<value_type> &summand : this->module_) {
+    if (summand.get_dimension() == dimension) {
+      // not optimal and only works for Dynamic_multi_filtration
+      auto birthList = summand._get_birth_list();
+      auto deathList = summand._get_death_list();
+      std::pair<std::vector<std::vector<value_type>>, std::vector<std::vector<value_type>>> corner;
+      corner.first.resize(birthList.size());
+      corner.second.resize(deathList.size());
+      unsigned int i = 0;
+      for (auto &b : birthList) {
+        corner.first[i] = std::vector<value_type>(b[0].begin(), b[0].end());
+        ++i;
+      }
+      i = 0;
+      for (auto &d : deathList) {
+        corner.second[i] = std::vector<value_type>(d[0].begin(), d[0].end());
+        ++i;
+      }
+      list.push_back(std::move(corner));
+    }
+  }
+  return list;
+}
+
+template <typename value_type>
+std::vector<std::vector<std::pair<value_type, value_type>>> Module<value_type>::get_barcode2(
+    const Line<value_type> &l,
+    const dimension_type dimension) const {
+  constexpr const bool verbose = false;
+  std::vector<std::vector<std::pair<value_type, value_type>>> barcode(this->get_dimension() + 1);
+  for (auto i = 0; i < this->get_dimension(); ++i) {
+    barcode[i].reserve(this->size());
+  }
+  for (unsigned int i = 0; i < this->size(); i++) {
+    const Summand<value_type> &summand = this->module_[i];
+    if constexpr (verbose) std::cout << "Summand of dimension " << summand.get_dimension() << std::endl;
+
+    if (dimension != -1 && summand.get_dimension() != dimension) continue;
+    /* if (dimension != -1 && summand.get_dimension() > dimension) */
+    /* 	break; */
+    const auto &pushed_summand = summand.get_bar2(l);
+
+    barcode[summand.get_dimension()].push_back(pushed_summand);
+  }
+  return barcode;
+}
+
+template <typename value_type>
+MultiDiagram<multipers::tmp_interface::Filtration_value<value_type>, value_type>
+Module<value_type>::get_barcode(const Line<value_type> &l, const dimension_type dimension, const bool threshold) const {
+  using F = multipers::tmp_interface::Filtration_value<value_type>;
+  constexpr const bool verbose = false;
+  if constexpr (verbose)
+    std::cout << "Computing barcode of dimension " << dimension << " and threshold " << threshold << std::endl;
+  std::vector<MultiDiagram_point<F>> barcode(this->size());
+  std::pair<value_type, value_type> threshold_bounds;
+  if (threshold) threshold_bounds = l.get_bounds(this->box_);
+  unsigned int summand_idx = 0;
+  for (unsigned int i = 0; i < this->size(); i++) {
+    const Summand<value_type> &summand = this->module_[i];
+    if constexpr (verbose) std::cout << "Summand of dimension " << summand.get_dimension() << std::endl;
+
+    if (dimension != -1 && summand.get_dimension() != dimension) continue;
+    /* if (dimension != -1 && summand.get_dimension() > dimension) */
+    /* 	break; */
+    auto pushed_summand = summand.get_bar(l);
+
+    F &pbirth = pushed_summand.first;
+    F &pdeath = pushed_summand.second;
+    if constexpr (verbose) std::cout << "BAR : " << pbirth << " " << pdeath << std::endl;
+    if (threshold) {
+      auto min = l[threshold_bounds.first];
+      auto max = l[threshold_bounds.second];
+      if (!(pbirth < max) || !(pdeath > min)) {
+        /* continue; */  // We still need summands to be aligned. The price to
+        // pay is some memory.
+        pbirth = std::numeric_limits<F>::infinity();
+        pdeath = pbirth;
+      }
+      pbirth.push_to_least_common_upper_bound(min);
+      pdeath.pull_to_greatest_common_lower_bound(max);
+    }
+    barcode[summand_idx++] = MultiDiagram_point(summand.get_dimension(), pbirth, pdeath);
+  }
+  barcode.resize(summand_idx);
+  return MultiDiagram<F, value_type>(barcode);
+}
+
+template <typename value_type>
+MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> Module<value_type>::get_barcodes(
+    const std::vector<Line<value_type>> &lines,
+    const dimension_type dimension,
+    const bool threshold) const {
+  unsigned int nlines = lines.size();
+  MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> out(nlines);
+  tbb::parallel_for(0U, nlines, [&](unsigned int i) {
+    const Line<value_type> &l = lines[i];
+    out[i] = this->get_barcode(l, dimension, threshold);
+  });
+  return out;
+}
+
+template <typename value_type>
+std::vector<std::vector<std::vector<std::pair<value_type, value_type>>>> Module<value_type>::get_barcodes2(
+    const std::vector<Line<value_type>> &lines,
+    const dimension_type dimension) const {
+  unsigned int nlines = lines.size();
+  std::vector<std::vector<std::vector<std::pair<value_type, value_type>>>> out(
+      this->get_dimension() + 1, std::vector<std::vector<std::pair<value_type, value_type>>>(nlines));
+  tbb::parallel_for(0U, nlines, [&](unsigned int i) {
+    const Line<value_type> &l = lines[i];
+    for (const auto &summand : module_) {
+      if (dimension != -1 && summand.get_dimension() != dimension) continue;
+      const auto &bar = summand.get_bar2(l);
+      out[summand.get_dimension()][i].push_back(bar);
+    }
+  });
+  return out;
+}
+
+template <typename value_type>
+MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> Module<value_type>::get_barcodes(
+    const std::vector<typename Line<value_type>::Point_t> &basepoints,
+    const dimension_type dimension,
+    const bool threshold) const {
+  unsigned int nlines = basepoints.size();
+  MultiDiagrams<multipers::tmp_interface::Filtration_value<value_type>, value_type> out(nlines);
+  // for (unsigned int i = 0; i < nlines; i++){
+  tbb::parallel_for(0U, nlines, [&](unsigned int i) {
+    const Line<value_type> &l = Line<value_type>(basepoints[i]);
+    out[i] = this->get_barcode(l, dimension, threshold);
+  });
+  return out;
+}
+
+template <typename value_type>
+std::vector<int> Module<value_type>::euler_curve(
+    const std::vector<multipers::tmp_interface::Filtration_value<value_type>> &points) const {
+  unsigned int npts = points.size();
+  std::vector<int> out(npts);
+  // #pragma omp parallel for
+  tbb::parallel_for(0U, static_cast<unsigned int>(out.size()), [&](unsigned int i) {
+    auto &euler_char = out[i];
+    const auto &point = points[i];
+    /* #pragma omp parallel for reduction(+ : euler_char) */
+    for (const Summand<value_type> &I : this->module_) {
+      if (I.contains(point)) {
+        int sign = I.get_dimension() % 2 ? -1 : 1;
+        euler_char += sign;
+      }
+    }
+  });
+  return out;
+}
+
+template <typename value_type>
+inline Box<value_type> Module<value_type>::get_bounds() const {
+  dimension_type num_parameters = box_.get_lower_corner().size();
+  typename Box<value_type>::Point_t lower_bound(num_parameters, std::numeric_limits<value_type>::infinity());
+  typename Box<value_type>::Point_t upper_bound(num_parameters, -std::numeric_limits<value_type>::infinity());
+  for (const auto &summand : module_) {
+    const auto &summand_bounds = summand.get_bounds();
+    const auto &[m, M] = summand_bounds.get_bounding_corners();
+    for (auto parameter = 0; parameter < num_parameters; parameter++) {
+      lower_bound[parameter] = std::min(m[parameter], lower_bound[parameter]);
+      upper_bound[parameter] = std::min(M[parameter], upper_bound[parameter]);
+    }
+  }
+  return Box(lower_bound, upper_bound);
+}
+
+template <typename value_type>
+inline void Module<value_type>::rescale(const std::vector<value_type> &rescale_factors, int degree) {
+  for (auto &summand : module_) {
+    if (degree == -1 or summand.get_dimension() == degree) summand.rescale(rescale_factors);
+  }
+}
+
+template <typename value_type>
+inline void Module<value_type>::translate(const std::vector<value_type> &translation, int degree) {
+  for (auto &summand : module_) {
+    if (degree == -1 or summand.get_dimension() == degree) summand.translate(translation);
+  }
+}
+
+template <typename value_type>
+inline std::vector<value_type> Module<value_type>::compute_pixels_of_degree(
+    const typename module_type::iterator start,
+    const typename module_type::iterator end,
+    const value_type delta,
+    const value_type p,
+    const bool normalize,
+    const Box<value_type> &box,
+    const std::vector<std::vector<value_type>> &coordinates,
+    const int n_jobs) {
+  unsigned int num_pixels = coordinates.size();
+  std::vector<value_type> out(num_pixels);
+  value_type moduleWeight = 0;
+  {  // for Timer
+    Debug::Timer timer("Computing module weight ...", verbose);
+    for (auto it = start; it != end; it++)  //  precomputes interleaving restricted to box for all summands.
+      it->get_interleaving(box);
+    if (p == 0) {
+      // #pragma omp parallel for reduction(+ : moduleWeight)
+      for (auto it = start; it != end; it++) {
+        moduleWeight += it->get_interleaving() > 0;
+      }
+    } else if (p != inf) {
+      // #pragma omp parallel for reduction(+ : moduleWeight)
+      for (auto it = start; it != end; it++) {
+        // /!\ TODO deal with inf summands (for the moment,  depends on the box
+        // ...)
+        if (it->get_interleaving() > 0 && it->get_interleaving() != inf)
+          moduleWeight += std::pow(it->get_interleaving(), p);
+      }
+    } else {
+      // #pragma omp parallel for reduction(std::max : moduleWeight)
+      for (auto it = start; it != end; it++) {
+        if (it->get_interleaving() > 0 && it->get_interleaving() != inf)
+          moduleWeight = std::max(moduleWeight, it->get_interleaving());
+      }
+    }
+  }  // Timer death
+  if (verbose) std::cout << "Module " << start->get_dimension() << " has weight : " << moduleWeight << "\n";
+  if (!moduleWeight) return out;
+
+  if constexpr (Debug::debug)
+    if (moduleWeight < 0) {
+      if constexpr (Debug::debug) std::cout << "!! Negative weight !!" << std::endl;
+      // 		image.clear();
+      return {};
+    }
+
+  oneapi::tbb::task_arena arena(n_jobs);  // limits the number of threads
+  arena.execute([&] {
+    tbb::parallel_for(0u, num_pixels, [&](unsigned int i) {
+      out[i] = _get_pixel_value(start, end, coordinates[i], delta, p, normalize, moduleWeight);
+    });
+  });
+  return out;
+}
+
+template <typename value_type>
+inline Module<int64_t> Module<value_type>::grid_squeeze(const std::vector<std::vector<value_type>> &grid) const {
+  auto dimension = this->get_dimension();
+  Module<int64_t> out(this->size());
+  for (auto i = 0u; i < this->size(); ++i) {
+    const auto &interval = this->operator[](i);
+    out[i] = interval.grid_squeeze(grid);
+  }
+  return out;
+}
+
+template <typename value_type>
+inline Summand<int64_t> Summand<value_type>::grid_squeeze(const std::vector<std::vector<value_type>> &grid) const {
+  auto dimension = this->get_dimension();
+  Summand<int64_t> out(
+      compute_coordinates_in_grid(birth_corners_, grid), compute_coordinates_in_grid(death_corners_, grid), dimension_);
+  return out;
+}
+
+/**
+ * dim, summand, (birth/death), num_pts, num_parameters
+ */
+template <typename value_type>
+inline typename Module<value_type>::idx_dump_type Module<value_type>::to_idx(
+    const std::vector<std::vector<value_type>> &grid) const {
+  unsigned int num_parameters = grid.size();
+  auto dimension = this->get_dimension();
+  idx_dump_type out(dimension + 1);
+  for (auto i = 0u; i < this->size(); ++i) {
+    auto &interval = this->operator[](i);
+    auto &out_of_dim = out[interval.get_dimension()];
+    out_of_dim.reserve(this->size());
+    std::pair<std::vector<std::vector<int>>, std::vector<std::vector<int>>> interval_idx;
+
+    auto &birth_idx = interval_idx.first;
+    birth_idx.reserve(interval._get_birth_list().size());
+    auto &death_idx = interval_idx.second;
+    death_idx.reserve(interval._get_death_list().size());
+
+    for (const auto &pt : interval._get_birth_list()) {
+      std::vector<int> pt_idx(pt.size());
+      for (auto i = 0u; i < num_parameters; ++i) {
+        pt_idx[i] = std::distance(grid[i].begin(), std::lower_bound(grid[i].begin(), grid[i].end(), pt[i]));
+      }
+      birth_idx.push_back(pt_idx);
+    }
+    for (const auto &pt : interval._get_death_list()) {
+      std::vector<int> pt_idx(pt.size());
+      for (auto i = 0u; i < num_parameters; ++i) {
+        pt_idx[i] = std::distance(grid[i].begin(), std::lower_bound(grid[i].begin(), grid[i].end(), pt[i]));
+      }
+      death_idx.push_back(pt_idx);
+    }
+    out_of_dim.push_back(interval_idx);
+  }
+  return out;
+}
+
+template <typename value_type>
+std::vector<int> inline to_grid_coord(const typename Summand<value_type>::filtration_type &pt,
+                                      const std::vector<std::vector<value_type>> &grid) {
+  std::size_t num_parameters = grid.size();
+  std::vector<int> out(num_parameters);
+  if (pt.is_plus_inf() || pt.is_nan()) [[unlikely]] {
+    for (size_t i = 0; i < num_parameters; ++i) out[i] = grid[i].size() - 1;
+    return out;
+  }
+  if (pt.is_minus_inf()) [[unlikely]] {
+    for (size_t i = 0; i < num_parameters; ++i) out[i] = 0;
+    return out;
+  }
+  // pt has to be of size num_parameters now
+  for (size_t i = 0u; i < num_parameters; ++i) {
+    if (pt(0, i) >= grid[i].back()) [[unlikely]]
+      out[i] = grid[i].size() - 1;
+    else if (pt(0, i) <= grid[i][0]) [[unlikely]] {
+      out[i] = 0;
+    } else {
+      auto temp = std::distance(grid[i].begin(), std::lower_bound(grid[i].begin(), grid[i].end(), pt(0, i)));
+      if (std::abs(grid[i][temp] - pt(0, i)) < std::abs(grid[i][temp - 1] - pt(0, i))) {
+        out[i] = temp;
+      } else {
+        out[i] = temp - 1;
+      }
+    }
+  }
+  return out;
+}
+
+template <typename value_type>
+std::vector<std::vector<std::vector<int>>> inline Module<value_type>::to_flat_idx(
+    const std::vector<std::vector<value_type>> &grid) const {
+  std::vector<std::vector<std::vector<int>>> out(3);
+  auto &idx = out[0];
+  auto &births = out[1];
+  auto &deaths = out[2];
+
+  idx.resize(2);
+  idx[0].resize(this->size());
+  idx[1].resize(this->size());
+
+  // some heuristic: usually
+  births.reserve(2 * this->size());
+  deaths.reserve(2 * this->size());
+  for (auto i = 0u; i < this->size(); ++i) {
+    auto &interval = this->operator[](i);
+    idx[0][i] = interval._get_birth_list().size();
+    for (const auto &pt : interval._get_birth_list()) {
+      births.push_back(to_grid_coord(pt, grid));
+    }
+    idx[1][i] = interval._get_death_list().size();
+    for (const auto &pt : interval._get_death_list()) {
+      deaths.push_back(to_grid_coord(pt, grid));
+    }
+  }
+  return out;
+}
+
+template <typename value_type>
+std::vector<int> inline Module<value_type>::get_degree_splits() const {
+  std::vector<int> splits = {};
+  int current_degree = 0;
+  for (auto i = 0u; i < this->size(); ++i) {
+    const auto &summand = this->operator[](i);
+    while (summand.get_dimension() > current_degree) [[unlikely]] {
+      current_degree++;
+      splits.push_back(i);
+    }
+  }
+  return splits;
+}
+
+template <typename value_type>
+inline void Module<value_type>::_compute_2D_image(Module::image_type &image,
+                                                  const typename module_type::iterator start,
+                                                  const typename module_type::iterator end,
+                                                  const value_type delta,
+                                                  const value_type p,
+                                                  const bool normalize,
+                                                  const Box<value_type> &box,
+                                                  const unsigned int horizontalResolution,
+                                                  const unsigned int verticalResolution) {
+  image.resize(horizontalResolution, std::vector<value_type>(verticalResolution));
+  value_type moduleWeight = 0;
+  {  // for Timer
+    Debug::Timer timer("Computing module weight ...", verbose);
+    for (auto it = start; it != end; it++)  //  precomputes interleaving restricted to box for all summands.
+      it->get_interleaving(box);
+    if (p == 0) {
+      /* #pragma omp parallel for reduction(+ : moduleWeight) */
+      for (auto it = start; it != end; it++) {
+        moduleWeight += it->get_interleaving() > 0;
+      }
+    } else if (p != inf) {
+      /* #pragma omp parallel for reduction(+ : moduleWeight) */
+      for (auto it = start; it != end; it++) {
+        // /!\ TODO deal with inf summands (for the moment,  depends on the box
+        // ...)
+        if (it->get_interleaving() > 0 && it->get_interleaving() != inf)
+          moduleWeight += std::pow(it->get_interleaving(), p);
+      }
+    } else {
+      /* #pragma omp parallel for reduction(std::max : moduleWeight) */
+      for (auto it = start; it != end; it++) {
+        if (it->get_interleaving() > 0 && it->get_interleaving() != inf)
+          moduleWeight = std::max(moduleWeight, it->get_interleaving());
+      }
+    }
+  }  // Timer death
+  if (verbose) std::cout << "Module " << start->get_dimension() << " has weight : " << moduleWeight << "\n";
+  if (!moduleWeight) return;
+
+  if constexpr (Debug::debug)
+    if (moduleWeight < 0) {
+      if constexpr (Debug::debug) std::cout << "!! Negative weight !!" << std::endl;
+      // 		image.clear();
+      return;
+    }
+
+  value_type stepX = (box.get_upper_corner()[0] - box.get_lower_corner()[0]) / horizontalResolution;
+  value_type stepY = (box.get_upper_corner()[1] - box.get_lower_corner()[1]) / verticalResolution;
+
+  {  // for Timer
+    Debug::Timer timer("Computing pixel values ...", verbose);
+
+    tbb::parallel_for(0U, horizontalResolution, [&](unsigned int i) {
+      tbb::parallel_for(0U, verticalResolution, [&](unsigned int j) {
+        image[i][j] = _get_pixel_value(start,
+                                       end,
+                                       {box.get_lower_corner()[0] + stepX * i, box.get_lower_corner()[1] + stepY * j},
+                                       delta,
+                                       p,
+                                       normalize,
+                                       moduleWeight);
+      });
+    });
+  }  // Timer death
+}
+
+template <typename value_type>
+inline void Module<value_type>::_compute_2D_image(Module::image_type &image,
+                                                  const typename module_type::const_iterator start,
+                                                  const typename module_type::const_iterator end,
+                                                  unsigned int horizontalResolution,
+                                                  unsigned int verticalResolution,
+                                                  get_2dpixel_value_function_type get_pixel_value) const {
+  image.resize(horizontalResolution, std::vector<value_type>(verticalResolution));
+  const Box<value_type> &box = this->box_;
+  value_type stepX = (box.get_upper_corner()[0] - box.get_lower_corner()[0]) / horizontalResolution;
+  value_type stepY = (box.get_upper_corner()[1] - box.get_lower_corner()[1]) / verticalResolution;
+
+  {  // for Timer
+    Debug::Timer timer("Computing pixel values ...", verbose);
+
+    // #pragma omp parallel for collapse(2)
+    // 			for (unsigned int i = 0; i < horizontalResolution; i++)
+    // 			{
+    // 				for (unsigned int j = 0; j < verticalResolution;
+    // j++)
+    // 				{
+    // 					image[i][j] = get_pixel_value(
+    // 						start,
+    // 						end,
+    // 						box.get_lower_corner()[0] +
+    // stepX * i,
+    // box.get_lower_corner()[1] + stepY * j);
+    // 				}
+    // 			}
+    tbb::parallel_for(0U, horizontalResolution, [&](unsigned int i) {
+      tbb::parallel_for(0U, verticalResolution, [&](unsigned int j) {
+        image[i][j] =
+            get_pixel_value(start, end, box.get_lower_corner()[0] + stepX * i, box.get_lower_corner()[1] + stepY * j);
+      });
+    });
+
+  }  // Timer death
+}
+
+template <typename value_type>
+inline value_type Module<value_type>::_get_pixel_value(const typename module_type::iterator start,
+                                                       const typename module_type::iterator end,
+                                                       const filtration_type x,
+                                                       const value_type delta,
+                                                       const value_type p,
+                                                       const bool normalize,
+                                                       const value_type moduleWeight) const {
+  value_type value = 0;
+  if (p == 0) {
+    /* #pragma omp parallel for reduction(+ : value) */
+    for (auto it = start; it != end; it++) {
+      value += it->get_local_weight(x, delta);
+    }
+    if (normalize) value /= moduleWeight;
+    return value;
+  }
+  if (p != inf) {
+    /* #pragma omp parallel for reduction(+ : value) */
+    for (auto it = start; it != end; it++) {
+      value_type summandWeight = it->get_interleaving();
+      value_type summandXWeight = it->get_local_weight(x, delta);
+      value += std::pow(summandWeight, p) * summandXWeight;
+    }
+    if (normalize) value /= moduleWeight;
+    return value;
+  }
+
+  /* #pragma omp parallel for reduction(std::max : value) */
+  for (auto it = start; it != end; it++) {
+    value = std::max(value, it->get_local_weight(x, delta));
+  }
+  return value;
+}
+
+/////////////////////////////////////////////////
+
+template <typename value_type>
+inline Summand<value_type>::Summand(int number_of_parameters)
+    : birth_corners_(number_of_parameters, births_type::T_inf),
+      death_corners_(number_of_parameters, -births_type::T_inf),
+      distanceTo0_(-1),
+      dimension_(-1) {}
+
+template <typename value_type>
+inline Summand<value_type>::Summand(const std::vector<value_type> &birth_corners,
+                                    const std::vector<value_type> &death_corners,
+                                    int num_parameters,
+                                    dimension_type dimension)
+    : birth_corners_(birth_corners.begin(), birth_corners.end(), num_parameters),
+      death_corners_(death_corners.begin(), death_corners.end(), num_parameters),
+      distanceTo0_(-1),
+      dimension_(dimension) {}
+
+template <typename value_type>
+inline bool Summand<value_type>::contains(const filtration_type &x) const {
+  bool out = false;
+  // only works with Dynamic_multi_filtration_value
+  for (const auto &birth : this->birth_corners_) {  // checks if there exists a birth smaller than x
+    if (birth <= x[0]) {
+      out = true;
+      break;
+    }
+  }
+  if (!out) return false;
+  out = false;
+  for (const auto &death : this->death_corners_) {
+    if (x[0] <= death) {
+      out = true;
+      break;
+    }
+  }
+  return out;
+}
+
+template <typename value_type>
+inline Summand<value_type>::Summand(const births_type &birth_corners,
+                                    const deaths_type &death_corners,
+                                    dimension_type dimension)
+    : birth_corners_(birth_corners), death_corners_(death_corners), distanceTo0_(-1), dimension_(dimension) {}
+
+template <typename value_type>
+inline value_type Summand<value_type>::get_interleaving(const Box<value_type> &box) {
+  _compute_interleaving(box);
+  return distanceTo0_;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::get_interleaving() const {
+  return distanceTo0_;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::get_local_weight(const filtration_type &x, const value_type delta) const {
+  bool rectangle = delta <= 0;
+
+  assert(x.num_parameters() == birth_corners_.num_parameters());
+
+  filtration_type mini(x.num_parameters());
+  filtration_type maxi(x.num_parameters());
+  if constexpr (Gudhi::multi_filtration::RangeTraits<filtration_type>::is_dynamic_multi_filtration) {
+    mini.force_generator_size_to_number_of_parameters(0);
+    maxi.force_generator_size_to_number_of_parameters(0);
+  }
+
+  // box on which to compute the local weight
+  for (unsigned int i = 0; i < x.num_parameters(); i++) {
+    mini(0, i) = delta <= 0 ? x(0, i) + delta : x(0, i) - delta;
+    maxi(0, i) = delta <= 0 ? x(0, i) - delta : x(0, i) + delta;
+  }
+
+  // Pre-allocating
+  std::vector<filtration_type> birthList(birth_corners_.num_generators());
+  std::vector<filtration_type> deathList(death_corners_.num_generators());
+  // Only works with Dynamic_multi_filtration
+  filtration_type birth(birth_corners_.num_parameters());
+  filtration_type death(birth_corners_.num_parameters());
+  unsigned int lastEntry = 0;
+  for (const auto &b : birth_corners_) {
+    birth[0] = b;
+    if (birth <= maxi) {
+      unsigned int dim = birth.num_parameters();
+      filtration_type tmpBirth(dim);
+      if constexpr (Gudhi::multi_filtration::RangeTraits<filtration_type>::is_dynamic_multi_filtration)
+        tmpBirth.force_generator_size_to_number_of_parameters(0);
+      for (unsigned int i = 0; i < dim; i++) {
+        tmpBirth(0, i) = std::max(birth(0, i), mini(0, i));
+      }
+
+      swap(birthList[lastEntry], tmpBirth);
+      lastEntry++;
+    }
+  }
+  birthList.resize(lastEntry);
+
+  // Thresholds birthlist & deathlist to B_inf(x,delta)
+  lastEntry = 0;
+  for (const auto &d : death_corners_) {
+    death[0] = d;
+    if (death >= mini) {
+      unsigned int dim = death.num_parameters();
+      filtration_type tmpDeath(dim);
+      if constexpr (Gudhi::multi_filtration::RangeTraits<filtration_type>::is_dynamic_multi_filtration)
+        tmpDeath.force_generator_size_to_number_of_parameters(0);
+      for (unsigned int i = 0; i < dim; i++) {
+        tmpDeath(0, i) = std::min(death(0, i), maxi(0, i));
+      }
+
+      swap(deathList[lastEntry], tmpDeath);
+      lastEntry++;
+    }
+  }
+  deathList.resize(lastEntry);
+
+  value_type local_weight = 0;
+  if (!rectangle) {
+    // Local weight is inteleaving to 0 of module restricted to the square
+    // #pragma omp parallel for reduction(std::max: local_weight)
+    Box<value_type> trivial_box;
+    for (const filtration_type &birth : birthList) {
+      if (birth.num_parameters() == 0) continue;
+      for (const filtration_type &death : deathList) {
+        if (death.num_parameters() > 0)
+          local_weight = std::max(local_weight,
+                                  _get_max_diagonal(birth,
+                                                    death,
+                                                    trivial_box));  // if box is empty, does not thredhold
+                                                                    // (already done before).
+      }
+    }
+    return local_weight / (2 * std::abs(delta));
+  } else {
+    // local weight is the volume of the largest rectangle in the restricted
+    // module #pragma omp parallel for reduction(std::max: local_weight)
+    for (const filtration_type &birth : birthList) {
+      if (birth.num_parameters() == 0) continue;
+      for (const filtration_type &death : deathList) {
+        if (death.num_parameters() > 0) local_weight = std::max(local_weight, _rectangle_volume(birth, death));
+      }
+    }
+    return local_weight / std::pow(2 * std::abs(delta), x.num_parameters());
+  }
+}
+
+template <typename value_type>
+inline std::tuple<int, int> Summand<value_type>::distance_idx_to_lower(const filtration_type &x) const {
+  value_type distance_to_lower = std::numeric_limits<value_type>::infinity();
+  int b_idx = -1;  // argmin_b max_i (b-x)_x
+  int param = 0;
+  auto count = 0u;
+  // Only works with Dynamic_multi_parameter_filtration
+  for (const auto &birth : birth_corners_) {
+    value_type temp = -std::numeric_limits<value_type>::infinity();  // max_i(birth - x)_+
+    int temp_idx = 0;
+    for (auto i = 0u; i < birth.size(); ++i) {
+      value_type plus = birth[i] - x(0, i);
+      if (plus > temp) {
+        temp_idx = i;
+        temp = plus;
+      }
+    }
+    if (temp < distance_to_lower) {
+      distance_to_lower = temp;
+      param = temp_idx;
+      b_idx = count;
+    }
+    ++count;
+  }
+  return {b_idx, param};
+}
+
+template <typename value_type>
+inline std::tuple<int, int> Summand<value_type>::distance_idx_to_upper(const filtration_type &x) const {
+  value_type distance_to_upper = std::numeric_limits<value_type>::infinity();
+  int d_idx = -1;  // argmin_d max_i (x-death)
+  int param = 0;
+  auto count = 0u;
+  // Only works with Dynamic_multi_parameter_filtration
+  for (const auto &death : death_corners_) {
+    value_type temp = -std::numeric_limits<value_type>::infinity();  // max_i(death-x)_+
+    int temp_idx = 0;
+    for (auto i = 0u; i < death.size(); ++i) {
+      value_type plus = x(0, i) - death[i];
+      if (plus > temp) {
+        temp_idx = i;
+        temp = plus;
+      }
+    }
+    if (temp < distance_to_upper) {
+      distance_to_upper = temp;
+      param = temp_idx;
+      d_idx = count;
+    }
+    ++count;
+  }
+  return {d_idx, param};
+}
+
+template <typename value_type>
+inline std::vector<int> Summand<value_type>::distance_idx_to(const filtration_type &x, bool full) const {
+  const auto &[a, b] = Summand::distance_idx_to_lower(x);
+  const auto &[c, d] = Summand::distance_idx_to_upper(x);
+  if (full) [[unlikely]]
+    return {a, b, c, d};
+  else {
+    return {a, c};
+  }
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::distance_to_lower(const filtration_type &x, bool negative) const {
+  value_type distance_to_lower = std::numeric_limits<value_type>::infinity();
+  for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+    value_type temp = negative ? -std::numeric_limits<value_type>::infinity() : 0;
+    for (unsigned int p = 0; p < birth_corners_.num_parameters(); ++p) {
+      temp = std::max(temp, birth_corners_(g, p) - x(0, p));
+    }
+    distance_to_lower = std::min(distance_to_lower, temp);
+  }
+  return distance_to_lower;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::distance_to_upper(const filtration_type &x, bool negative) const {
+  value_type distance_to_upper = std::numeric_limits<value_type>::infinity();
+  for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+    value_type temp = negative ? -std::numeric_limits<value_type>::infinity() : 0;
+    for (unsigned int p = 0; p < death_corners_.num_parameters(); ++p) {
+      temp = std::max(temp, x(0, p) - death_corners_(g, p));
+    }
+    distance_to_upper = std::min(distance_to_upper, temp);
+  }
+  return distance_to_upper;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::distance_to(const filtration_type &x, bool negative) const {
+  return std::max(Summand::distance_to_lower(x, negative), Summand::distance_to_upper(x, negative));
+}
+
+template <typename value_type>
+inline std::pair<value_type, value_type> Summand<value_type>::get_bar2(const Line<value_type> &l) const {
+  constexpr const bool verbose = false;
+  if constexpr (verbose)
+    std::cout << "Computing bar of this summand of dimension " << this->get_dimension() << std::endl;
+  value_type pushed_birth = std::numeric_limits<value_type>::infinity();
+  value_type pushed_death = -pushed_birth;
+  for (filtration_type birth : this->_get_birth_list()) {
+    value_type pb = l.compute_forward_intersection(birth);
+    pushed_birth = std::min(pb, pushed_birth);
+  }
+  //
+  for (const filtration_type &death : this->_get_death_list()) {
+    value_type pd = l.compute_backward_intersection(death);
+    pushed_death = std::max(pd, pushed_death);
+  }
+
+  if (!(pushed_birth <= pushed_death)) {
+    if constexpr (verbose) std::cout << "Birth <!= Death ! Ignoring this value" << std::endl;
+    return {inf, inf};
+  }
+  if constexpr (verbose) {
+    std::cout << "Final values" << pushed_birth << " ----- " << pushed_death << std::endl;
+  }
+  return {pushed_birth, pushed_death};
+}
+
+template <typename value_type>
+inline std::pair<multipers::tmp_interface::Filtration_value<value_type>,
+                 multipers::tmp_interface::Filtration_value<value_type>>
+Summand<value_type>::get_bar(const Line<value_type> &l) const {
+  using F = multipers::tmp_interface::Filtration_value<value_type>;
+  constexpr const bool verbose = false;
+  int num_param = birth_corners_.num_parameters();
+  if constexpr (verbose)
+    std::cout << "Computing bar of this summand of dimension " << this->get_dimension() << std::endl;
+  F pushed_birth = F::inf(num_param);
+  F pushed_death = F::minus_inf(num_param);
+  for (const F &birth : this->get_birth_list()) {
+    F pb = l[l.compute_forward_intersection(birth)];
+    if constexpr (verbose)
+      std::cout << "Updating birth " << pushed_birth << " with " << pb << " pushed at " << birth << " "
+                << pushed_birth.is_plus_inf();
+    if ((pb <= pushed_birth) || pushed_birth.is_plus_inf()) {
+      swap(pushed_birth, pb);
+      if constexpr (verbose) std::cout << " swapped !";
+    }
+    if constexpr (verbose) std::cout << std::endl;
+  }
+  //
+  for (const F &death : this->get_death_list()) {
+    F pd = l[l.compute_backward_intersection(death)];
+    if constexpr (verbose)
+      std::cout << "Updating death " << pushed_death << " with " << pd << " pushed at " << death << " "
+                << pushed_death.is_minus_inf() << pushed_death[0];
+    if ((pd >= pushed_death) || pushed_death.is_minus_inf()) {
+      swap(pushed_death, pd);
+      if constexpr (verbose) std::cout << " swapped !";
+    }
+    if constexpr (verbose) std::cout << std::endl;
+  }
+
+  if (!(pushed_birth <= pushed_death)) {
+    if constexpr (verbose) std::cout << "Birth <!= Death ! Ignoring this value" << std::endl;
+    return {F::inf(num_param), F::inf(num_param)};
+  }
+  if constexpr (verbose) {
+    std::cout << "Final values" << pushed_birth << " ----- " << pushed_death << std::endl;
+  }
+  return {pushed_birth, pushed_death};
+}
+
+/**
+ * @brief Adds the bar @p bar to the indicator module @p summand if @p bar
+ * is non-trivial (ie. not reduced to a point or, if @p threshold is true,
+ * its thresholded version should not be reduced to a point) .
+ *
+ * @param bar p_bar: to add to the support of the summand
+ * @param summand p_summand: indicator module which is being completed
+ * @param basepoint p_basepoint: basepoint of the line of the bar
+ * @param birth p_birth: birth container (for memory optimization purposes).
+ * Has to be of the size @p basepoint.size()+1.
+ * @param death p_death: death container. Same purpose as @p birth but for death.
+ * @param threshold p_threshold: If true, will threshold the bar with @p box.
+ * @param box p_box: Only useful if @p threshold is set to true.
+ */
+
+template <typename value_type>
+inline void Summand<value_type>::add_bar(value_type baseBirth,
+                                         value_type baseDeath,
+                                         const typename Line<value_type>::Point_t &basepoint,
+                                         filtration_type &birth,
+                                         filtration_type &death,
+                                         const bool threshold,
+                                         const Box<value_type> &box) {
+  // bar is trivial in that case
+  if (baseBirth >= baseDeath) return;
+  // #pragma omp simd
+  // 		for (unsigned int j = 0; j < birth.size() - 1; j++)
+  // 		{
+  // 			birth[j] = basepoint[j] + baseBirth;
+  // 			death[j] = basepoint[j] + baseDeath;
+  // 		}
+  // 		birth.back() = baseBirth;
+  // 		death.back() = baseDeath;
+
+  /* #pragma omp simd */
+  for (unsigned int j = 0; j < birth.num_parameters() - 1; j++) {
+    value_type temp = basepoint[j] + baseBirth;
+    // The box is assumed to contain all of the filtration values, if its
+    // outside, its inf.
+    birth(0, j) = temp < box.get_lower_corner()[j] ? negInf : temp;
+    temp = basepoint[j] + baseDeath;
+    death(0, j) = temp > box.get_upper_corner()[j] ? inf : temp;
+  }
+  birth(0, birth.num_parameters() - 1) = baseBirth < box.get_lower_corner().back() ? negInf : baseBirth;
+  death(0, birth.num_parameters() - 1) = baseDeath > box.get_upper_corner().back() ? inf : baseDeath;
+
+  if (threshold) {
+    // std::cout << box;
+    threshold_down(birth, box, basepoint);
+    threshold_up(death, box, basepoint);
+  }
+  _add_birth(birth);
+  _add_death(death);
+}
+
+template <typename value_type>
+inline void Summand<value_type>::add_bar(const filtration_type &birth, const filtration_type &death) {
+  _add_birth(birth);
+  _add_death(death);
+}
+
+template <typename value_type>
+inline void Summand<value_type>::add_bar(const typename Line<value_type>::Point_t &basepoint,
+                                         value_type birth,
+                                         value_type death,
+                                         const Box<value_type> &box) {
+  constexpr const bool verbose = false;
+  if (birth >= death) return;
+  if constexpr (verbose) {
+    std::cout << "Bar : " << basepoint + birth << "--" << basepoint + death << std::endl;
+  }
+  auto inf = std::numeric_limits<value_type>::infinity();
+  auto container = basepoint + birth;
+  for (auto i = 0u; i < container.size(); i++) {
+    if (container[i] < box.get_lower_corner()[i]) container[i] = -inf;
+  }
+  _add_birth(container);
+  container = basepoint + death;
+  for (auto i = 0u; i < container.size(); i++) {
+    if (container[i] > box.get_upper_corner()[i]) container[i] = inf;
+  }
+  _add_death(container);
+}
+
+template <typename value_type>
+inline std::vector<typename Summand<value_type>::filtration_type> Summand<value_type>::_get_birth_list() const {
+  std::vector<filtration_type> res(birth_corners_.num_generators(), filtration_type(birth_corners_.num_parameters()));
+  for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+    // could be done in a more generic way, but this should be the fastest without changing the current interface
+    if constexpr (Gudhi::multi_filtration::RangeTraits<filtration_type>::is_dynamic_multi_filtration)
+      res[g].force_generator_size_to_number_of_parameters(0);
+    for (unsigned int p = 0; p < birth_corners_.num_parameters(); ++p) {
+      res[g](0, p) = birth_corners_(g, p);
+    }
+  }
+
+  return res;
+}
+
+template <typename value_type>
+inline std::vector<typename Summand<value_type>::filtration_type> Summand<value_type>::_get_death_list() const {
+  // TODO: factorize with _get_birth_list
+  std::vector<filtration_type> res(death_corners_.num_generators(), filtration_type(death_corners_.num_parameters()));
+  for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+    // could be done in a more generic way, but this should be the fastest without changing the current interface
+    if constexpr (Gudhi::multi_filtration::RangeTraits<filtration_type>::is_dynamic_multi_filtration)
+      res[g].force_generator_size_to_number_of_parameters(0);
+    for (unsigned int p = 0; p < death_corners_.num_parameters(); ++p) {
+      res[g](0, p) = death_corners_(g, p);
+    }
+  }
+
+  return res;
+}
+
+template <typename value_type>
+inline std::vector<multipers::tmp_interface::Filtration_value<value_type>> Summand<value_type>::get_birth_list() const {
+  using F = multipers::tmp_interface::Filtration_value<value_type>;
+  std::vector<F> res(birth_corners_.num_generators(), F(birth_corners_.num_parameters()));
+  for (unsigned int g = 0; g < birth_corners_.num_generators(); ++g) {
+    // could be done in a more generic way, but this should be the fastest without changing the current interface
+    if constexpr (Gudhi::multi_filtration::RangeTraits<F>::is_dynamic_multi_filtration)
+      res[g].force_generator_size_to_number_of_parameters(0);
+    for (unsigned int p = 0; p < birth_corners_.num_parameters(); ++p) {
+      res[g](0, p) = birth_corners_(g, p);
+    }
+  }
+
+  return res;
+}
+
+template <typename value_type>
+inline std::vector<multipers::tmp_interface::Filtration_value<value_type>> Summand<value_type>::get_death_list() const {
+  // TODO: factorize with _get_birth_list
+  using F = multipers::tmp_interface::Filtration_value<value_type>;
+  std::vector<F> res(death_corners_.num_generators(), F(death_corners_.num_parameters()));
+  for (unsigned int g = 0; g < death_corners_.num_generators(); ++g) {
+    // could be done in a more generic way, but this should be the fastest without changing the current interface
+    if constexpr (Gudhi::multi_filtration::RangeTraits<F>::is_dynamic_multi_filtration)
+      res[g].force_generator_size_to_number_of_parameters(0);
+    for (unsigned int p = 0; p < death_corners_.num_parameters(); ++p) {
+      res[g](0, p) = death_corners_(g, p);
+    }
+  }
+
+  return res;
+}
+
+template <typename value_type>
+const Summand<value_type>::births_type &Summand<value_type>::get_upset() const {
+  return birth_corners_;
+}
+
+template <typename value_type>
+const Summand<value_type>::deaths_type &Summand<value_type>::get_downset() const {
+  return death_corners_;
+};
+
+template <typename value_type>
+inline void Summand<value_type>::clean() {
+  // birth_corners_.erase(std::remove_if(birth_corners_.begin(),
+  //                                     birth_corners_.end(),
+  //                                     [](const std::vector<value_type> &bp) {
+  //                                       // return std::any_of(
+  //                                       //     bp.begin(), bp.end(),
+  //                                       //     [](float value) { return !std::isfinite(value); });
+  //                                       bp.size() == 0;
+  //                                     }),
+  //                      birth_corners_.end());
+  // birth_corners_.simplify();
+  // TODO : clean
+}
+
+template <typename value_type>
+inline void Summand<value_type>::complete_birth(const value_type precision) {
+  if (!birth_corners_.is_finite()) return;
+
+  for (std::size_t i = 0; i < birth_corners_.num_generators(); i++) {
+    for (std::size_t j = i + 1; j < birth_corners_.num_generators(); j++) {
+      value_type dinf = d_inf(birth_corners_[i], birth_corners_[j]);
+      if (dinf < .99 * precision) {  // for machine error ?
+        _factorize_min(birth_corners_[i], birth_corners_[j]);
+        birth_corners_[j] = std::remove_reference_t<decltype(birth_corners_[j])>::inf();
+        i++;
+      }
+    }
+  }
+  birth_corners_.simplify();
+  // _clean(birth_corners_);
+}
+
+template <typename value_type>
+inline void Summand<value_type>::complete_death(const value_type precision) {
+  if (!death_corners_.is_finite()) return;
+
+  for (std::size_t i = 0; i < death_corners_.num_generators(); i++) {
+    for (std::size_t j = i + 1; j < death_corners_.num_generators(); j++) {
+      value_type d = d_inf(death_corners_[i], death_corners_[j]);
+      if (d < .99 * precision) {
+        _factorize_max(death_corners_[i], death_corners_[j]);
+        death_corners_[j] = std::remove_reference_t<decltype(death_corners_[j])>::minus_inf();
+        i++;
+      }
+    }
+  }
+
+  death_corners_.simplify();
+  // _clean(death_corners_);
+}
+
+template <typename value_type>
+inline dimension_type Summand<value_type>::get_dimension() const {
+  return dimension_;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::get_landscape_value(const std::vector<value_type> &x) const {
+  value_type out = 0;
+  Box<value_type> trivial_box;
+  // Only works with Dynamic_multi_parameter_filtration and is not optimal
+  filtration_type b(birth_corners_.num_parameters());
+  filtration_type d(death_corners_.num_parameters());
+  for (const auto &birth : this->birth_corners_) {
+    for (const auto &death : this->death_corners_) {
+      b[0] = birth;
+      d[0] = death;
+      value_type value =
+          std::min(this->_get_max_diagonal(b, x, trivial_box), this->_get_max_diagonal(x, d, trivial_box));
+      out = std::max(out, value);
+    }
+  }
+  return out;
+}
+
+template <typename value_type>
+inline void Summand<value_type>::set_dimension(dimension_type dimension) {
+  dimension_ = dimension;
+}
+
+template <typename value_type>
+inline void Summand<value_type>::_compute_interleaving(const Box<value_type> &box) {
+  distanceTo0_ = 0;
+  /* #pragma omp parallel for reduction(max : distanceTo0_) */
+  // Only works with Dynamic_multi_parameter_filtration and is not optimal
+  filtration_type b(birth_corners_.num_parameters());
+  filtration_type d(death_corners_.num_parameters());
+  for (const auto &birth : birth_corners_) {
+    for (const auto &death : death_corners_) {
+      b[0] = birth;
+      d[0] = death;
+      distanceTo0_ = std::max(distanceTo0_, _get_max_diagonal(b, d, box));
+    }
+  }
+}
+
+/**
+ * @brief Adds @p birth to the summand's @p birth_list if it is not induced
+ * from the @p birth_list (ie. not comparable or smaller than another birth),
+ * and removes unnecessary birthpoints (ie. birthpoints that are induced
+ * by @p birth).
+ *
+ * @param birth_list p_birth_list: birthpoint list of a summand
+ * @param birth p_birth: birth to add to the summand
+ */
+
+template <typename value_type>
+inline void Summand<value_type>::_add_birth(const filtration_type &birth) {
+  birth_corners_.add_generator(birth[0]);
+  return;
+
+  // // TODO : DEPRECATE THIS OLD CODE
+  // if (birth_corners_.empty()) {
+  //   birth_corners_.push_back(birth);
+  //   return;
+  // }
+
+  // for (const auto &current_birth : birth_corners_) {
+  //   if (birth >= current_birth) {
+  //     return;
+  //   }
+  // }
+  // // this birth value is useful, we can now remove useless other filtrations
+  // for (auto &current_birth : birth_corners_) {
+  //   if ((!current_birth.empty()) && (birth <= current_birth)) {
+  //     current_birth.clear();
+  //   }
+  // }
+
+  // _clean(birth_corners_);
+  // birth_corners_.push_back(birth);
+}
+
+/**
+ * @brief Adds @p death to the summand's @p death_list if it is not induced
+ * from the @p death_list (ie. not comparable or greater than another death),
+ * and removes unnecessary deathpoints (ie. deathpoints that are induced
+ * by @p death)
+ *
+ * @param death_list p_death_list: List of deathpoints of a summand
+ * @param death p_death: deathpoint to add to this list
+ */
+
+template <typename value_type>
+inline void Summand<value_type>::_add_death(const filtration_type &death) {
+  death_corners_.add_generator(death[0]);
+  return;
+  // // TODO:  Deprecate this old code
+  // if (death_corners_.empty()) {
+  //   death_corners_.push_back(death);
+  //   return;
+  // }
+
+  // for (const auto &current_death : death_corners_) {
+  //   if (death <= current_death) {
+  //     return;
+  //   }
+  // }
+  // // this death value is useful, we can now remove useless other filtrations
+  // for (auto &current_death : death_corners_) {
+  //   if (!current_death.empty() && (death >= current_death)) {
+  //     current_death.clear();
+  //   }
+  // }
+  // _clean(death_corners_);
+  // death_corners_.push_back(death);
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::_get_max_diagonal(const filtration_type &birth,
+                                                         const filtration_type &death,
+                                                         const Box<value_type> &box) const {
+  // assumes birth and death to be never NaN
+  assert(birth.num_parameters() == death.num_parameters() && "Inputs must be of the same size !");
+  assert((box.is_trivial() || birth.num_parameters() == box.get_dimension()) && "Inputs must be of the same size !");
+
+  value_type s = inf;
+  bool threshold_flag = !box.is_trivial();
+  if (threshold_flag) {
+    for (unsigned int i = 0; i < birth.num_parameters(); ++i) {
+      value_type max_i = box.get_upper_corner()[i];
+      value_type min_i = box.get_lower_corner()[i];
+      value_type t_death = death.is_plus_inf() ? max_i : (death.is_minus_inf() ? -inf : std::min(death(0, i), max_i));
+      value_type t_birth = birth.is_plus_inf() ? inf : (birth.is_minus_inf() ? min_i : std::max(birth(0, i), min_i));
+      s = std::min(s, t_death - t_birth);
+    }
+  } else {
+    for (unsigned int i = 0; i < birth.num_parameters(); i++) {
+      s = std::min(s, death(0, i) - birth(0, i));
+    }
+  }
+
+  return s;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::_rectangle_volume(const filtration_type &a, const filtration_type &b) const {
+  if constexpr (Debug::debug) assert(a.num_parameters() == b.num_parameters() && "Inputs must be of the same size !");
+  value_type s = b(0, 0) - a(0, 0);
+  for (unsigned int i = 1; i < a.num_parameters(); i++) {
+    s = s * (b(0, i) - a(0, i));
+  }
+  return s;
+}
+
+template <typename value_type>
+inline value_type Summand<value_type>::d_inf(const Generator &a, const Generator &b) const {
+  if (a.size() == 0 || b.size() == 0 || a.size() != b.size()) return inf;
+
+  value_type d = std::abs(a[0] - b[0]);
+  for (unsigned int i = 1; i < a.size(); i++) d = std::max(d, std::abs(a[i] - b[i]));
+
+  return d;
+}
+
+template <typename value_type>
+inline void Summand<value_type>::_factorize_min(Generator &a, const Generator &b) {
+  /* if (Debug::debug && (a.empty() || b.empty())) */
+  /* { */
+  /* 	std::cout << "Empty corners ??\n"; */
+  /* 	return; */
+  /* } */
+
+  for (unsigned int i = 0; i < std::min(b.size(), a.size()); i++) a[i] = std::min(a[i], b[i]);
+}
+
+template <typename value_type>
+inline void Summand<value_type>::_factorize_max(Generator &a, const Generator &b) {
+  /* if (Debug::debug && (a.empty() || b.empty())) */
+  /* { */
+  /* 	std::cout << "Empty corners ??\n"; */
+  /* 	return; */
+  /* } */
+
+  for (unsigned int i = 0; i < std::min(b.size(), a.size()); i++) a[i] = std::max(a[i], b[i]);
+}
+
+/**
+ * @brief Cleans empty entries of a corner list
+ *
+ * @param list corner list to clean
+ * @param keep_sort If true, will keep the order of the corners,
+ * with a computational overhead. Defaults to false.
+ */
+// WARNING Does permute the output.
+
+template <typename value_type>
+inline void Summand<value_type>::_clean(std::vector<filtration_type> &list, bool keep_inf) {
+  list.erase(std::remove_if(list.begin(),
+                            list.end(),
+                            [keep_inf](const filtration_type &a) {
+                              return a.num_parameters() == 0 || ((!keep_inf) && (a.is_plus_inf() || a.is_minus_inf()));
+                            }),
+             list.end());
+}
+
+template <typename value_type>
+Module<value_type> direct_sum(const Module<value_type> &a, const Module<value_type> &b) {
+  Module<value_type> out;
+  out.module_.resize(a.size() + b.size());
+  std::copy(a.module_.begin(), a.module_.end(), out.module_.begin());
+  std::copy(b.module_.begin(), b.module_.end(), out.module_.begin() + a.size());
+  out.box_ = get_smallest_enclosing_box(a.box_, b.box_);
+  return out;
+}
+
+template <typename value_type>
+Module<value_type> direct_sum(std::vector<Module<value_type> *> &modules) {
+  Module<value_type> out;
+  size_t total_size = 0;
+  for (const auto &m : modules) {
+    total_size += m->size();
+  }
+  out.module_.resize(total_size);
+  size_t index = 0;
+  for (const auto &m : modules) {
+    std::copy(m->module_.begin(), m->module_.end(), out.module_.begin() + index);
+    index += m->size();
+  }
+  return out;
+}
+}  // namespace mma
+}  // namespace multiparameter
+}  // namespace Gudhi
+
+#endif  // APPR