multipers 2.3.3b6__cp311-cp311-manylinux_2_39_x86_64.whl

multipers/multi_parameter_rank_invariant/persistence_slices.h

@@ -0,0 +1,148 @@
+#pragma once
+
+#include "gudhi/Flag_complex_edge_collapser.h"
+#include <gudhi/Persistent_cohomology.h>
+#include <gudhi/Simplex_tree_multi.h>
+#include <gudhi/Multi_persistence/Box.h>
+
+namespace Gudhi {
+namespace multiparameter {
+
+struct Simplex_tree_float {  // smaller simplextrees
+  typedef linear_indexing_tag Indexing_tag;
+  typedef std::int32_t Vertex_handle;
+  typedef float Filtration_value;
+  typedef std::uint32_t Simplex_key;
+  static const bool store_key = true;
+  static const bool store_filtration = true;
+  static const bool contiguous_vertices = false;  // TODO OPTIMIZATION : maybe make the simplextree contiguous when
+                                                  // calling grid_squeeze ?
+  static const bool link_nodes_by_label = true;
+  static const bool stable_simplex_handles = false;
+  static const bool is_multi_parameter = false;
+};
+
+// using Simplex_tree_float = Simplex_tree_options_fast_persistence;
+
+using Gudhi::multi_persistence::Box;
+using Simplex_tree_std = Simplex_tree<Simplex_tree_float>;
+
+using Barcode = std::vector<std::pair<Simplex_tree_std::Filtration_value, Simplex_tree_std::Filtration_value>>;
+
+inline Barcode compute_dgm(Simplex_tree_std &st, int degree) {
+  st.initialize_filtration(true);
+  constexpr int coeff_field_characteristic = 11;
+  constexpr Simplex_tree_std::Filtration_value min_persistence = 0;
+  bool persistence_dim_max = st.dimension() == degree;
+  Gudhi::persistent_cohomology::Persistent_cohomology<Simplex_tree_std, Gudhi::persistent_cohomology::Field_Zp> pcoh(
+      st, persistence_dim_max);
+  pcoh.init_coefficients(coeff_field_characteristic);
+  pcoh.compute_persistent_cohomology(min_persistence);
+  const auto &persistent_pairs = pcoh.intervals_in_dimension(degree);
+  if constexpr (false) {
+    std::cout << "Number of bars : " << persistent_pairs.size() << "\n";
+  }
+  return persistent_pairs;
+}
+
+template <typename degree_type, class interface_std_like>
+inline std::vector<Barcode> compute_dgms(interface_std_like &st,
+                                         const std::vector<degree_type> &degrees,
+                                         int num_collapses,
+                                         int expansion_dim) {
+  std::vector<Barcode> out(degrees.size());
+  static_assert(!interface_std_like::Options::is_multi_parameter,
+                "Can only compute persistence for 1-parameter simplextrees.");
+  const bool verbose = false;
+  if (num_collapses > 0) {
+    auto collapsed_st = collapse_edges(st, num_collapses);
+    return compute_dgms(collapsed_st, degrees, 0, expansion_dim);
+  }
+  // if (expansion_dim == -1){
+  // 	int max_dim = *std::max_element(degrees.begin(), degrees.end()) +1;
+  // 	st.expansion(max_dim);
+  // }
+  else if (expansion_dim > 0) {
+    st.expansion(expansion_dim);
+  }
+
+  st.initialize_filtration(true);  // true is ignore_infinite_values
+  constexpr int coeff_field_characteristic = 11;
+  constexpr typename interface_std_like::Filtration_value min_persistence = 0;
+
+  bool persistence_dim_max = false;
+  for (auto degree : degrees) {
+    if (st.dimension() == degree) {
+      persistence_dim_max = true;
+      break;
+    }
+  }
+  //
+  if constexpr (verbose) {
+    std::cout << "Computing dgm of st:\n";
+    for (auto &sh : st.filtration_simplex_range()) {
+      std::cout << "dim: " << st.dimension(sh) << " vertices: ";
+      for (auto v : st.simplex_vertex_range(sh)) {
+        std::cout << v << " ";
+      }
+      std::cout << " filtration: ";
+      std::cout << st.filtration(sh) << "\n";
+    }
+    return out;
+  }
+
+  Gudhi::persistent_cohomology::Persistent_cohomology<interface_std_like, Gudhi::persistent_cohomology::Field_Zp> pcoh(
+      st, persistence_dim_max);
+  pcoh.init_coefficients(coeff_field_characteristic);
+  pcoh.compute_persistent_cohomology(min_persistence);
+  for (auto i = 0u; i < degrees.size(); i++) {
+    out[i] = pcoh.intervals_in_dimension(degrees[i]);
+  }
+  return out;
+}
+
+// small wrapper
+template <typename degree_type, class interface_std_like>
+inline std::vector<Barcode> compute_dgms(interface_std_like &st,
+                                         const std::vector<degree_type> &degrees,
+                                         int expand_collapse_dim = 0) {
+  if (expand_collapse_dim > 0) return compute_dgms(st, degrees, 10, expand_collapse_dim);
+  return compute_dgms(st, degrees, 0, 0);
+}
+
+// Adapted version from the Simplextree interface
+template <class simplextree_like>
+inline simplextree_like collapse_edges(simplextree_like &st, int num_collapses) {
+  using Filtered_edge = std::tuple<typename simplextree_like::Vertex_handle,
+                                   typename simplextree_like::Vertex_handle,
+                                   typename simplextree_like::Filtration_value>;
+  std::vector<Filtered_edge> edges;
+  for (auto sh : st.skeleton_simplex_range(1)) {
+    if (st.dimension(sh) == 1) {
+      const auto filtration = st.filtration(sh);
+      if (filtration == std::numeric_limits<typename simplextree_like::Filtration_value>::infinity()) continue;
+      typename simplextree_like::Simplex_vertex_range rg = st.simplex_vertex_range(sh);
+      auto vit = rg.begin();
+      typename simplextree_like::Vertex_handle v = *vit;
+      typename simplextree_like::Vertex_handle w = *++vit;
+      edges.emplace_back(v, w, filtration);
+    }
+  }
+  for (int iteration = 0; iteration < num_collapses; iteration++) {
+    auto current_size = edges.size();
+    edges = Gudhi::collapse::flag_complex_collapse_edges(std::move(edges));
+    if (edges.size() >= current_size) break;  // no need to do more
+  }
+  simplextree_like collapsed_stree;
+  // Copy the original 0-skeleton
+  for (auto sh : st.skeleton_simplex_range(0)) {
+    collapsed_stree.insert_simplex({*(st.simplex_vertex_range(sh).begin())}, st.filtration(sh));
+  }
+  // Insert remaining edges
+  for (auto [x, y, filtration] : edges) {
+    collapsed_stree.insert_simplex({x, y}, filtration);
+  }
+  return collapsed_stree;
+}
+
+}}  // namespace Gudhi::multiparameter

multipers/multi_parameter_rank_invariant/rank_invariant.h

@@ -0,0 +1,369 @@
+#pragma once
+#include <gudhi/Simplex_tree_multi.h>
+#include "gudhi/truc.h"
+#include "multi_parameter_rank_invariant/persistence_slices.h"
+#include "tensor/tensor.h"
+#include <algorithm>
+#include <cstddef>
+#include <gudhi/One_critical_filtration.h>
+#include <oneapi/tbb/enumerable_thread_specific.h>
+#include <oneapi/tbb/parallel_for.h>
+#include <ostream>
+#include <utility>  // std::pair
+#include <vector>
+
+namespace Gudhi {
+namespace multiparameter {
+namespace rank_invariant { 
+using Index = truc_interface::index_type;
+
+// using Elbow = std::vector<std::pair<>>;grid
+template <typename index_type>
+inline void push_in_elbow(index_type &i, index_type &j, const index_type I, const index_type J) {
+  if (j < J) {
+    j++;
+    return;
+  }
+  if (i < I) {
+    i++;
+    return;
+  }
+  j++;
+  return;
+}
+
+template <typename index_type, typename value_type>
+inline value_type get_slice_rank_filtration(const value_type x,
+                                            const value_type y,
+                                            const index_type I,
+                                            const index_type J) {
+  if (x > static_cast<value_type>(I))
+    return std::numeric_limits<value_type>::has_infinity ? std::numeric_limits<value_type>::infinity()
+                                                         : std::numeric_limits<value_type>::max();
+  if (y > static_cast<value_type>(J)) return I + static_cast<index_type>(y);
+  return J + static_cast<index_type>(x);
+}
+
+template <typename index_type>
+inline std::pair<index_type, index_type> get_coordinates(index_type in_slice_value, index_type I, index_type J) {
+  if (in_slice_value <= J) return {0, J};
+  if (in_slice_value <= I + J) return {in_slice_value - J, J};
+  return {I, in_slice_value - I};
+}
+
+template <typename dtype, typename index_type, typename Filtration>
+inline void compute_2d_rank_invariant_of_elbow(
+    Simplex_tree<Gudhi::multi_persistence::Simplex_tree_options_multidimensional_filtration<Filtration>> &st_multi,
+    Simplex_tree_std &_st_container,                           // copy of st_multi
+    const tensor::static_tensor_view<dtype, index_type> &out,  // assumes its a zero tensor
+    const index_type I,
+    const index_type J,
+    const std::vector<index_type> &grid_shape,
+    const std::vector<index_type> &degrees,
+    const int expand_collapse_max_dim = 0) {
+  // const bool verbose = false; // verbose
+  using value_type = typename Simplex_tree_std::Filtration_value;
+  constexpr const value_type inf = std::numeric_limits<value_type>::infinity();
+
+  // const auto X = grid_shape[1],  Y = grid_shape[2]; // First axis is degree
+  const auto Y = grid_shape[2];
+  // Fills the filtration in the container
+  // TODO : C++23 zip, when Apples clang will stop being obsolete
+  auto sh_standard = _st_container.complex_simplex_range().begin();
+  auto _end = _st_container.complex_simplex_range().end();
+  auto sh_multi = st_multi.complex_simplex_range().begin();
+  for (; sh_standard != _end; ++sh_multi, ++sh_standard) {
+    const Filtration &multi_filtration = st_multi.filtration(*sh_multi);
+    value_type filtration_in_slice;
+    if constexpr (Filtration::is_multi_critical) {
+      filtration_in_slice = inf;
+      for (const auto &stuff : multi_filtration) {
+        value_type x = stuff[0];
+        value_type y = stuff[1];
+        filtration_in_slice = std::min(filtration_in_slice, get_slice_rank_filtration(x, y, I, J));
+      }
+    } else {
+      value_type x = multi_filtration[0];
+      value_type y = multi_filtration[1];
+      filtration_in_slice = get_slice_rank_filtration(x, y, I, J);
+    }
+    _st_container.assign_filtration(*sh_standard, filtration_in_slice);
+  }
+  const std::vector<Barcode> &barcodes = compute_dgms(_st_container, degrees, expand_collapse_max_dim);
+  index_type degree_index = 0;
+  for (const auto &barcode : barcodes) {  // TODO range view cartesian product
+    for (const auto &bar : barcode) {
+      auto birth = static_cast<index_type>(bar.first);
+      auto death = static_cast<index_type>(
+          std::min(bar.second,
+                   static_cast<typename Simplex_tree_std::Filtration_value>(Y + I)));  // I,J atteints, pas X ni Y
+
+      // todo : optimize
+      // auto [a,b] = get_coordinates(birth, I,J);
+      for (auto intermediate_birth = birth; intermediate_birth < death; intermediate_birth++) {
+        for (auto intermediate_death = intermediate_birth; intermediate_death < death; intermediate_death++) {
+          auto [i, j] = get_coordinates(intermediate_birth, I, J);
+          auto [k, l] = get_coordinates(intermediate_death, I, J);
+          if (((i < k || j == J) && (j < l || k == I))) {
+            // std::vector<index_type> coordinates_to_remove =
+            // {degree_index,i,j,k,l}; out[coordinates_to_remove]++;
+            out[{degree_index, i, j, k, l}]++;
+          }
+        }
+      }
+    }
+    degree_index++;
+  }
+}
+
+template <typename dtype, typename index_type, typename Filtration>
+inline void compute_2d_rank_invariant(
+    Simplex_tree<Gudhi::multi_persistence::Simplex_tree_options_multidimensional_filtration<Filtration>> &st_multi,
+    const tensor::static_tensor_view<dtype, index_type> &out,  // assumes its a zero tensor
+    const std::vector<index_type> &grid_shape,
+    const std::vector<index_type> &degrees,
+    bool expand_collapse) {
+  if (degrees.size() == 0) return;
+  assert(st_multi.get_number_of_parameters() == 2);
+  Simplex_tree_std st_;
+  Gudhi::multi_persistence::flatten(st_, st_multi,
+                                    0);  // copies the st_multi to a standard 1-pers simplextree
+  const int max_dim = expand_collapse ? *std::max_element(degrees.begin(), degrees.end()) + 1 : 0;
+  index_type X = grid_shape[1];
+  index_type Y = grid_shape[2];                                                // First axis is degree
+  tbb::enumerable_thread_specific<Simplex_tree_std> thread_simplex_tree(st_);  // initialize with a good simplextree
+  tbb::parallel_for(0, X, [&](index_type I) {
+    tbb::parallel_for(0, Y, [&](index_type J) {
+      auto &st_container = thread_simplex_tree.local();
+      compute_2d_rank_invariant_of_elbow<dtype, index_type, Filtration>(
+          st_multi, st_container, out, I, J, grid_shape, degrees, max_dim);
+    });
+  });
+}
+
+template <typename Filtration, typename dtype, typename indices_type, typename... Args>
+void compute_rank_invariant_python(
+    Simplex_tree<Gudhi::multi_persistence::Simplex_tree_options_multidimensional_filtration<Filtration>> &st_multi,
+    dtype *data_ptr,
+    const std::vector<indices_type> grid_shape,
+    const std::vector<indices_type> degrees,
+    indices_type n_jobs,
+    bool expand_collapse) {
+  if (degrees.size() == 0) return;
+  tensor::static_tensor_view<dtype, indices_type> container(data_ptr, grid_shape);  // assumes its a zero tensor
+
+  oneapi::tbb::task_arena arena(n_jobs);  // limits the number of threads
+  arena.execute([&] {
+    compute_2d_rank_invariant<dtype, indices_type, Filtration>(
+        st_multi, container, grid_shape, degrees, expand_collapse);
+  });
+
+  return;
+}
+
+template <class PersBackend,
+          class Structure,
+          class MultiFiltration = Gudhi::multi_filtration::One_critical_filtration<float>,
+          typename dtype,
+          typename index_type>
+inline void compute_2d_rank_invariant_of_elbow(
+    typename truc_interface::Truc<PersBackend, Structure, MultiFiltration>::ThreadSafe &slicer,  // truc slicer
+    const tensor::static_tensor_view<dtype, index_type> &out,                               // assumes its a zero tensor
+    const index_type I,
+    const index_type J,
+    const std::vector<index_type> &grid_shape,
+    const std::vector<index_type> &degrees,
+    // std::vector<Index> &order_container,                                 // constant size
+    // std::vector<typename MultiFiltration::value_type> &one_persistence,  // constant size
+    const bool flip_death = false,
+    const bool ignore_inf = true) {
+  using value_type = typename MultiFiltration::value_type;
+  const auto &filtrations_values = slicer.get_filtrations();
+  auto num_generators = filtrations_values.size();
+  // one_persistence.resize(num_generators); // local variable should be
+  // initialized correctly
+  const auto Y = grid_shape[2];
+  constexpr const bool verbose = false;
+  if constexpr (verbose) std::cout << "filtration_in_slice : [ ";
+  for (auto i = 0u; i < num_generators; ++i) {
+    const auto &f = filtrations_values[i];
+    value_type filtration_in_slice = MultiFiltration::Generator::T_inf;
+    if constexpr (MultiFiltration::is_multi_critical) {
+      for (const auto &stuff : f) {
+        value_type x = stuff[0];
+        value_type y = stuff[1];
+
+        filtration_in_slice = std::min(filtration_in_slice, get_slice_rank_filtration(x, y, I, J));
+      }
+    } else {
+      value_type x = f[0];
+      value_type y = f[1];
+      filtration_in_slice = get_slice_rank_filtration(x, y, I, J);
+    }
+    if constexpr (verbose) std::cout << filtration_in_slice << ",";
+    slicer.get_one_filtration()[i] = filtration_in_slice;
+  }
+  if constexpr (verbose) std::cout << "\b]" << std::endl;
+
+  index_type degree_index = 0;
+  // order_container.resize(slicer.num_generators()); // local variable should
+  // be initialized correctly
+  // TODO : use slicer::ThreadSafe instead of maintaining one_pers & order
+  // BUG : This will break as soon as slicer interface change
+
+  using bc_type = typename truc_interface::Truc<PersBackend, Structure, MultiFiltration>::split_barcode;
+  bc_type barcodes;
+  if constexpr (PersBackend::is_vine) {
+    // slicer.set_one_filtration(one_persistence);
+    if (I == 0 && J == 0) [[unlikely]]  // this is dangerous, assumes it starts at 0 0
+    {
+      // TODO : This is a good optimization but needs a patch on
+      // PersistenceMatrix std::vector<bool>
+      // degrees_index(slicer.get_dimensions().back()+1, false); for (const auto
+      // &degree : degrees) {
+      //   if (degree <= slicer.get_dimensions())
+      //     degrees_index[degree] = true;
+      // }
+      // slicer.compute_persistence(degrees_index);
+      slicer.compute_persistence();
+    } else {
+      slicer.vineyard_update();
+    }
+    barcodes = slicer.get_barcode();
+  } else {
+    slicer.compute_persistence(ignore_inf);
+    barcodes = slicer.get_barcode();
+  }
+
+  // note one_pers not necesary when vine, but does the same computation
+
+  for (auto degree : degrees) {
+    // this assumes barcodes degrees starts from 0
+    if constexpr (verbose) std::cout << "Adding Barcode of degree " << degree << std::endl;
+    if (degree >= static_cast<index_type>(barcodes.size())) continue;
+    const auto &barcode = barcodes[degree];
+    for (const auto &bar : barcode) {
+      if (bar.first > Y + I) continue;
+      if constexpr (verbose)
+        std::cout << bar.first << " " << bar.second << "checkinf: " << MultiFiltration::Generator::T_inf << " ==? "
+                  << (bar.first == MultiFiltration::Generator::T_inf) << std::endl;
+      auto birth = static_cast<index_type>(bar.first);
+      auto death = static_cast<index_type>(
+          std::min(bar.second,
+                   static_cast<typename MultiFiltration::value_type>(Y + I)));  // I,J atteints, pas X ni Y
+      if constexpr (false) std::cout << "Birth " << birth << " Death " << death << std::endl;
+      for (auto intermediate_birth = birth; intermediate_birth < death; intermediate_birth++) {
+        for (auto intermediate_death = intermediate_birth; intermediate_death < death; intermediate_death++) {
+          auto [i, j] = get_coordinates(intermediate_birth, I, J);
+          auto [k, l] = get_coordinates(intermediate_death, I, J);
+          if (((i < k || j == J) && (j < l || k == I))) {
+            if (flip_death)
+              out[{degree_index, i, j, I - 1 - k, J - 1 - l}]++;
+            else
+              out[{degree_index, i, j, k, l}]++;
+          }
+          if constexpr (false) std::cout << degree_index << " " << i << " " << j << " " << k << " " << l << std::endl;
+        }
+      }
+    }
+    degree_index++;
+  }
+};
+
+template <class PersBackend,
+          class Structure,
+          class MultiFiltration = Gudhi::multi_filtration::One_critical_filtration<float>,
+          typename dtype,
+          typename index_type>
+inline void compute_2d_rank_invariant(
+    truc_interface::Truc<PersBackend, Structure, MultiFiltration> &slicer,
+    const tensor::static_tensor_view<dtype, index_type> &out,  // assumes its a zero tensor
+    const std::vector<index_type> &grid_shape,
+    const std::vector<index_type> &degrees,
+    const bool flip_death,
+    const bool ignore_inf) {
+  if (degrees.size() == 0) return;
+  index_type X = grid_shape[1];
+  index_type Y = grid_shape[2];  // First axis is degree
+  constexpr const bool verbose = false;
+  if constexpr (verbose)
+    std::cout << "Shape " << grid_shape[0] << " " << grid_shape[1] << " " << grid_shape[2] << " " << grid_shape[3]
+              << " " << grid_shape[4] << std::endl;
+
+  using ThreadSafe = typename truc_interface::Truc<PersBackend, Structure, MultiFiltration>::ThreadSafe;
+  ThreadSafe slicer_thread(slicer);
+  tbb::enumerable_thread_specific<ThreadSafe> thread_locals(slicer_thread);
+  tbb::parallel_for(0, X, [&](index_type I) {
+    tbb::parallel_for(0, Y, [&](index_type J) {
+      if constexpr (verbose) std::cout << "Computing elbow " << I << " " << J << "...";
+      ThreadSafe &slicer = thread_locals.local();
+      compute_2d_rank_invariant_of_elbow<PersBackend, Structure, MultiFiltration, dtype, index_type>(
+          slicer, out, I, J, grid_shape, degrees, flip_death, ignore_inf);
+      if constexpr (verbose) std::cout << "Done!" << std::endl;
+    });
+  });
+}
+
+template <class PersBackend,
+          class Structure,
+          class MultiFiltration = Gudhi::multi_filtration::One_critical_filtration<float>,
+          typename dtype,
+          typename indices_type>
+void compute_rank_invariant_python(truc_interface::Truc<PersBackend, Structure, MultiFiltration> slicer,
+                                   dtype *data_ptr,
+                                   const std::vector<indices_type> grid_shape,
+                                   const std::vector<indices_type> degrees,
+                                   indices_type n_jobs,
+                                   const bool ignore_inf) {
+  if (degrees.size() == 0) return;
+  tensor::static_tensor_view<dtype, indices_type> container(data_ptr, grid_shape);  // assumes its a zero tensor
+  if constexpr (false){
+    std::cout  << "ignore_inf " << ignore_inf << std::endl;
+  }
+
+  oneapi::tbb::task_arena arena(PersBackend::is_vine ? 1 : n_jobs);  // limits the number of threads
+  arena.execute([&] { compute_2d_rank_invariant(slicer, container, grid_shape, degrees, false, ignore_inf); });
+
+  return;
+}
+
+template <typename PersBackend,
+          typename Structure,
+          typename MultiFiltration,
+          typename dtype = int,
+          typename indices_type = int>
+std::pair<std::vector<std::vector<indices_type>>, std::vector<dtype>> compute_rank_signed_measure(
+    truc_interface::Truc<PersBackend, Structure, MultiFiltration> slicer,
+    dtype *data_ptr,
+    const std::vector<indices_type> grid_shape,
+    const std::vector<indices_type> degrees,
+    indices_type n_jobs,
+    bool verbose,
+    const bool ignore_inf) {
+  if (degrees.size() == 0) return {{}, {}};
+  tensor::static_tensor_view<dtype, indices_type> container(data_ptr, grid_shape);  // assumes its a zero tensor
+  oneapi::tbb::task_arena arena(n_jobs);                                            // limits the number of threads
+  constexpr bool flip_death = true;
+  arena.execute([&] { compute_2d_rank_invariant(slicer, container, grid_shape, degrees, flip_death, ignore_inf); });
+
+  if (verbose) {
+    std::cout << "Done.\n";
+    std::cout << "Computing mobius inversion ..." << std::flush;
+  }
+
+  // for (indices_type axis :
+  // std::views::iota(2,st_multi.get_number_of_parameters()+1)) // +1 for the
+  // degree in axis 0
+  for (std::size_t axis = 0u; axis < slicer.num_parameters() + 1; axis++) container.differentiate(axis);
+  if (verbose) {
+    std::cout << "Done.\n";
+    std::cout << "Sparsifying the measure ..." << std::flush;
+  }
+  auto raw_signed_measure = container.sparsify({false, false, true, true});
+  if (verbose) {
+    std::cout << "Done.\n";
+  }
+  return raw_signed_measure;
+}
+
+}}}  // namespace Gudhi::multiparameter::rank_invariant

multipers/multiparameter_edge_collapse.py

@@ -0,0 +1,41 @@
+from tqdm import tqdm
+
+
+def _collapse_edge_list(
+    edges,
+    num: int = 0,
+    full: bool = False,
+    strong: bool = False,
+    progress: bool = False,
+):
+    """
+    Given an edge list defining a 1 critical 2 parameter 1 dimensional simplicial complex, simplificates this filtered simplicial complex, using filtration-domination's edge collapser.
+    """
+    from filtration_domination import (
+        remove_filtration_dominated,
+        remove_strongly_filtration_dominated,
+    )
+
+    n = len(edges)
+    if full:
+        num = 100
+    with tqdm(
+        range(num), total=num, desc="Removing edges", disable=not (progress)
+    ) as I:
+        for i in I:
+            if strong:
+                edges = remove_strongly_filtration_dominated(edges)  # nogil ?
+            else:
+                edges = remove_filtration_dominated(edges)
+            # Prevents doing useless collapses
+            if len(edges) >= n:
+                if full and strong:
+                    strong = False
+                    n = len(edges)
+                    # n = edges.size() # len(edges)
+                else:
+                    break
+            else:
+                n = len(edges)
+                # n = edges.size()
+    return edges