Farthest points (from Siargey)

git-svn-id: svn+ssh://scm.gforge.inria.fr/svnroot/gudhi/branches/subsampling_and_spatialsearching@1230 636b058d-ea47-450e-bf9e-a15bfbe3eedb

Former-commit-id: 3093bfd41948fecf5f6ccb2d1d77546c12e628f4

4 files changed, 307 insertions, 0 deletions

diff --git a/src/Subsampling/include/gudhi/Landmark_choice_by_farthest_point.h b/src/Subsampling/include/gudhi/Landmark_choice_by_farthest_point.h
new file mode 100644
index 00000000..198c9f9f
--- /dev/null
+++ b/src/Subsampling/include/gudhi/Landmark_choice_by_farthest_point.h
@@ -0,0 +1,158 @@
+/*    This file is part of the Gudhi Library. The Gudhi library
+ *    (Geometric Understanding in Higher Dimensions) is a generic C++
+ *    library for computational topology.
+ *
+ *    Author(s):       Siargey Kachanovich
+ *
+ *    Copyright (C) 2015  INRIA Sophia Antipolis-Méditerranée (France)
+ *
+ *    This program is free software: you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation, either version 3 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *    GNU General Public License for more details.
+ *
+ *    You should have received a copy of the GNU General Public License
+ *    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef LANDMARK_CHOICE_BY_FARTHEST_POINT_H_
+#define LANDMARK_CHOICE_BY_FARTHEST_POINT_H_
+
+#include <gudhi/Spatial_tree_data_structure.h>
+
+#include <iterator>
+#include <algorithm>  // for sort
+#include <vector>
+#include <random>
+#include <boost/heap/fibonacci_heap.hpp>
+
+namespace Gudhi {
+
+
+  template < typename Point_d,
+             typename Heap,
+             typename Tree,
+             typename Presence_table >
+  void update_heap( Point_d &l,
+                    unsigned nbL,
+                    Heap &heap,
+                    Tree &tree,
+                    Presence_table &table)
+  {
+    auto search = tree.query_incremental_ANN(l);
+    for (auto w: search) {
+      if (table[w.first].first)
+        if (w.second < table[w.first].second->second) {
+          heap.update(table[w.first].second, w);
+        }
+    }
+  }
+  
+  /** 
+   *  \ingroup witness_complex
+   *  \brief Landmark choice strategy by iteratively adding the farthest witness from the
+   *  current landmark set as the new landmark. 
+   *  \details It chooses nbL landmarks from a random access range `points` and
+   *  writes {witness}*{closest landmarks} matrix in `knn`.
+   *
+   *  The type KNearestNeighbors can be seen as 
+   *  Witness_range<Closest_landmark_range<Vertex_handle>>, where
+   *  Witness_range and Closest_landmark_range are random access ranges 
+   *  
+   */
+
+  template < typename Kernel,
+             typename Point_container,
+             typename OutputIterator>
+  void landmark_choice_by_farthest_point( Kernel& k,
+                                          Point_container const &points,
+                                          int nbL,
+                                          OutputIterator output_it)
+  {
+
+    // typedef typename Kernel::FT FT;
+    // typedef std::pair<unsigned, FT> Heap_node;
+    
+    // struct R_max_compare
+    // {
+    //   bool operator()(const Heap_node &rmh1, const Heap_node &rmh2) const
+    //   {
+    //     return rmh1.second < rmh2.second;
+    //   }
+    // };
+    
+    // typedef boost::heap::fibonacci_heap<Heap_node, boost::heap::compare<R_max_compare>> Heap;
+    // typedef Spatial_tree_data_structure<Kernel, Point_container> Tree;
+    // typedef std::vector< std::pair<bool, Heap_node*> > Presence_table;
+
+    typename Kernel::Squared_distance_d sqdist = k.squared_distance_d_object();
+    
+  //   Tree tree(points);
+  //   Heap heap;
+  //   Presence_table table(points.size());
+  //   for (auto p: table)
+  //     std::cout << p.first << "\n";
+  //   int number_landmarks = 0; // number of treated landmarks
+
+  //   double curr_max_dist = 0;                                      // used for defining the furhest point from L
+  //   const double infty = std::numeric_limits<double>::infinity();  // infinity (see next entry)
+  //   std::vector< double > dist_to_L(points.size(), infty);         // vector of current distances to L from points
+    
+  //   // Choose randomly the first landmark 
+  //   std::random_device rd;
+  //   std::mt19937 gen(rd());
+  //   std::uniform_int_distribution<> dis(1, 6);
+  //   int curr_landmark = dis(gen);
+    
+  //   do {
+  //     *output_landmarks++ = points[curr_landmark];
+  //     std::cout << curr_landmark << "\n";
+  //     number_landmarks++;
+  //   }
+  //   while (number_landmarks < nbL);
+  // }
+
+    int nb_points = boost::size(points);
+    assert(nb_points >= nbL);
+
+    int current_number_of_landmarks = 0;  // counter for landmarks
+    double curr_max_dist = 0;  // used for defining the furhest point from L
+    const double infty = std::numeric_limits<double>::infinity();  // infinity (see next entry)
+    std::vector< double > dist_to_L(nb_points, infty);  // vector of current distances to L from points
+
+    // Choose randomly the first landmark 
+    std::random_device rd;
+    std::mt19937 gen(rd());
+    std::uniform_int_distribution<> dis(1, 6);
+    int curr_max_w = dis(gen);
+
+    
+    for (current_number_of_landmarks = 0; current_number_of_landmarks != nbL; current_number_of_landmarks++) {
+      // curr_max_w at this point is the next landmark
+      *output_it++ = points[curr_max_w];
+      std::cout << curr_max_w << "\n";
+      unsigned i = 0;
+      for (auto& p : points) {
+        double curr_dist = sqdist(p, *(std::begin(points) + curr_max_w));
+        if (curr_dist < dist_to_L[i])
+          dist_to_L[i] = curr_dist;
+        ++i;
+      }
+      // choose the next curr_max_w
+      curr_max_dist = 0;
+      for (i = 0; i < dist_to_L.size(); i++)
+        if (dist_to_L[i] > curr_max_dist) {
+          curr_max_dist = dist_to_L[i];
+          curr_max_w = i;
+        }
+    }
+  }
+  
+}  // namespace Gudhi
+
+#endif  // LANDMARK_CHOICE_BY_FARTHEST_POINT_H_
diff --git a/src/Subsampling/include/gudhi/Landmark_choice_by_random_point.h b/src/Subsampling/include/gudhi/Landmark_choice_by_random_point.h
new file mode 100644
index 00000000..daa05d1a
--- /dev/null
+++ b/src/Subsampling/include/gudhi/Landmark_choice_by_random_point.h
@@ -0,0 +1,80 @@
+/*    This file is part of the Gudhi Library. The Gudhi library
+ *    (Geometric Understanding in Higher Dimensions) is a generic C++
+ *    library for computational topology.
+ *
+ *    Author(s):       Siargey Kachanovich
+ *
+ *    Copyright (C) 2016  INRIA Sophia Antipolis-Méditerranée (France)
+ *
+ *    This program is free software: you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation, either version 3 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *    GNU General Public License for more details.
+ *
+ *    You should have received a copy of the GNU General Public License
+ *    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef LANDMARK_CHOICE_BY_RANDOM_POINT_H_
+#define LANDMARK_CHOICE_BY_RANDOM_POINT_H_
+
+#include <boost/range/size.hpp>
+
+#include <random>     // random_device, mt19937
+#include <algorithm>  // shuffle
+#include <numeric>    // iota
+#include <iterator>
+#include <gudhi/Clock.h>
+
+
+namespace Gudhi {
+
+  /**
+   *  \ingroup witness_complex
+   * \brief Landmark choice strategy by taking random vertices for landmarks.
+   *
+   *  \details It chooses nbL distinct landmarks from a random access range `points`
+   *  and outputs them to an output iterator.
+   *  Point_container::iterator should be ValueSwappable and RandomAccessIterator.
+   */
+
+  template <typename Point_container,
+            typename OutputIterator>
+  void landmark_choice_by_random_point(Point_container const &points,
+                                       unsigned nbL,
+                                       OutputIterator output_it) {
+#ifdef GUDHI_LM_PROFILING
+    Gudhi::Clock t;
+#endif
+
+    unsigned nbP = boost::size(points);
+    assert(nbP >= nbL);
+    std::vector<int> landmarks(nbP);
+    std::iota(landmarks.begin(), landmarks.end(), 0);
+
+    std::random_device rd;
+    std::mt19937 g(rd());
+ 
+    std::shuffle(landmarks.begin(), landmarks.end(), g);
+    landmarks.resize(nbL);
+
+    for (int l: landmarks)
+      *output_it++ = points[l];
+    
+#ifdef GUDHI_LM_PROFILING
+      t.end();
+      std::cerr << "Random landmark choice took " << t.num_seconds()
+        << " seconds." << std::endl;
+#endif
+
+    
+  }
+
+}  // namespace Gudhi
+
+#endif  // LANDMARK_CHOICE_BY_RANDOM_POINT_H_
diff --git a/src/Subsampling/test/CMakeLists.txt b/src/Subsampling/test/CMakeLists.txt
new file mode 100644
index 00000000..3a45c685
--- /dev/null
+++ b/src/Subsampling/test/CMakeLists.txt
@@ -0,0 +1,22 @@
+cmake_minimum_required(VERSION 2.6)
+project(GUDHILandmarkingTest)
+
+# Landmarking test
+if(CGAL_FOUND)
+  if (NOT CGAL_VERSION VERSION_LESS 4.8.0)
+    message(STATUS "CGAL version: ${CGAL_VERSION}.")
+
+    find_package(Eigen3 3.1.0)
+    if (EIGEN3_FOUND)
+      message(STATUS "Eigen3 version: ${EIGEN3_VERSION}.")
+      include( ${EIGEN3_USE_FILE} )
+      include_directories (BEFORE "../../include")
+
+      add_executable( landmarking_UT landmarking.cpp ) 
+    else()
+      message(WARNING "Eigen3 not found. Version 3.1.0 is required for Landmarking feature.")
+    endif()
+  else()
+    message(WARNING "CGAL version: ${CGAL_VERSION} is too old to compile Landmarking feature. Version 4.8.0 is required.")
+  endif ()
+endif()
diff --git a/src/Subsampling/test/landmarking.cpp b/src/Subsampling/test/landmarking.cpp
new file mode 100644
index 00000000..3131c798
--- /dev/null
+++ b/src/Subsampling/test/landmarking.cpp
@@ -0,0 +1,47 @@
+// #ifdef _DEBUG
+// # define TBB_USE_THREADING_TOOL
+// #endif
+
+#include <gudhi/Landmark_choice_by_random_point.h>
+#include <gudhi/Landmark_choice_by_farthest_point.h>
+#include <vector>
+#include <iterator>
+
+#include <CGAL/Epick_d.h>
+
+
+int main() {
+  typedef CGAL::Epick_d<CGAL::Dynamic_dimension_tag>                K;
+  typedef typename K::FT                                            FT;
+  typedef typename K::Point_d                                       Point_d;
+  
+  std::vector<Point_d> vect;
+  vect.push_back(Point_d(std::vector<FT>({0,0,0,0})));
+  vect.push_back(Point_d(std::vector<FT>({0,0,0,1})));
+  vect.push_back(Point_d(std::vector<FT>({0,0,1,0})));
+  vect.push_back(Point_d(std::vector<FT>({0,0,1,1})));
+  vect.push_back(Point_d(std::vector<FT>({0,1,0,0})));
+  vect.push_back(Point_d(std::vector<FT>({0,1,0,1})));
+  vect.push_back(Point_d(std::vector<FT>({0,1,1,0})));
+  vect.push_back(Point_d(std::vector<FT>({0,1,1,1})));
+  vect.push_back(Point_d(std::vector<FT>({1,0,0,0})));
+  vect.push_back(Point_d(std::vector<FT>({1,0,0,1})));
+  vect.push_back(Point_d(std::vector<FT>({1,0,1,0})));
+  vect.push_back(Point_d(std::vector<FT>({1,0,1,1})));
+  vect.push_back(Point_d(std::vector<FT>({1,1,0,0})));
+  vect.push_back(Point_d(std::vector<FT>({1,1,0,1})));
+  vect.push_back(Point_d(std::vector<FT>({1,1,1,0})));
+  vect.push_back(Point_d(std::vector<FT>({1,1,1,1})));
+  
+  
+  std::vector<Point_d> landmarks;
+  Gudhi::landmark_choice_by_random_point(vect, 5, std::back_inserter(landmarks));
+  std::cout << "landmark vector contains: ";
+  for (auto l: landmarks)
+    std::cout << l << "\n";
+
+  landmarks.clear();
+  K k;
+  Gudhi::landmark_choice_by_farthest_point(k, vect, 16, std::back_inserter(landmarks));
+  
+}