Fix for Windows and Mac

build_extension.py

@@ -25,6 +25,7 @@ def build_extensions(self) -> None:
         # with clang++. This works across compilers (ignored by MSVC).
         for extension in self.extensions:
             extension.extra_compile_args.append("-std=c++11")
+            extension.extra_compile_args.append("-DLEMON_ONLY_TEMPLATES")
 
         try:
             build_ext.build_extensions(self)
@@ -34,6 +35,7 @@ def build_extensions(self) -> None:
             # so the code can compile on macOS with Anaconda.
             for extension in self.extensions:
                 extension.extra_compile_args.append("-stdlib=libc++")
+                extension.extra_compile_args.append("-DLEMON_ONLY_TEMPLATES")
                 extension.extra_link_args.append("-stdlib=libc++")
             build_ext.build_extensions(self)
 

fair_kmeans/_core.cpp

@@ -33,7 +33,9 @@ void postprocessData(double *array, double *sampleWeights, int *colors, uint n,
     int j = 0;
     for (uint i = 0; i < n * d; i += d)
     {
-        double w = sampleWeights[j];
+        // In a future version, we might consider using weights as doubles
+        // but currently the algorithm does not allow for that
+        int w = static_cast<int>(sampleWeights[j]);
         int c = colors[j];
         number[c] += 1;
         weightsum[c] += w;
@@ -109,7 +111,7 @@ extern "C"
                uint nc,
                uint maxIterations,
                double tolerance,
-               std::uint64_t seed,
+               int seed,
                int *labels,
                double *centers,
                bool updateCenters,

fair_kmeans/core.py

@@ -148,7 +148,7 @@ def _run_fair_clustering(
         c_k = ctypes.c_uint(self.n_clusters)
         c_n_colors = ctypes.c_uint(self.n_colors_)
 
-        c_random_state = ctypes.c_size_t(self._seed)
+        c_random_state = ctypes.c_int(self._seed)
 
         c_labels = (ctypes.c_int * n_samples)()
         c_centers = self.cluster_centers_.ctypes.data_as(

fair_kmeans/fair_clustering_tools.cpp

@@ -169,6 +169,12 @@ std::vector<Point> computeFairlets(const std::vector<std::vector<Point>> &colore
   return fairlets;
 }
 
+int randomDoubleToInt(double value, int min, int max)
+{
+  double result = (value * (max - min)) + min;
+  return static_cast<int>(std::ceil(result));
+}
+
 // Computes the fair assignment of points to a given set of centers. Returns new centroids.
 CentersValueChange compute_fair_assignment(
     std::vector<std::vector<Point>> &points,
@@ -336,18 +342,22 @@ CentersValueChange compute_fair_assignment(
   int centerindex = 0;
 
   std::mt19937 rng(seed); // random-number engine used (Mersenne-Twister in this case)
-  std::uniform_int_distribution<int> uni(0, points[0].size() - 1); // guaranteed unbiased
+  // Replaced this because it does not produce the same results on different platforms
+  //std::uniform_int_distribution<int> uni(0, points[0].size() - 1); // guaranteed unbiased
+  std::uniform_real_distribution<double> realDist(0.0, 1.0);
+
+  # define getRandomValue() randomDoubleToInt(realDist(rng), 0, points[0].size() - 1)
 
   // Choose k - 1 random integers between 0 and the size of the blue points
   // Such that we can use them in case our clusters are empty
   std::vector<Matching> matchingPairs(centers.size() - 1);
 
-  for (size_t i = 0; i < centers.size() - 1; ++i)
+  for (size_t i = 0; i < matchingPairs.size(); ++i)
   {
-    auto randomInteger = uni(rng);
+    auto randomInteger = getRandomValue();
     while (std::find_if(matchingPairs.begin(), matchingPairs.end(), find_id(randomInteger)) != matchingPairs.end())
     {
-      randomInteger = uni(rng);
+      randomInteger = getRandomValue();
     }
     matchingPairs[i].source = randomInteger;
   }

fair_kmeans/point.h

@@ -3,6 +3,8 @@
 #include <vector>
 #include <cstddef>
 
+typedef unsigned int uint;
+
 // Small class for storing weighted points
 class Point
 {

tests/datasets/adult-balanced_10.output

@@ -1,12 +1,12 @@
 1.27575e+10
-10
+8
 1
-56.1786,9.60714,0,859.607,32.75
+55.5714,9.35714,0,859.607,30.25
 39.5,12.5,99999,0,45
 47,10.75,8855.5,0,45
 41.8889,12.4444,12553,0,46.5
 36.9618,10.0632,3.23508,0,39.1671
-49.7,9.5,3841.2,0,31.6
+51.4,10.2,3841.2,0,38.6
 46.3667,10.8,0,2022.5,45.1
 42.625,12.5,61495.9,0,39.875
 38.6818,10.6364,5793.09,0,44

tests/datasets/adult-balanced_100.output

@@ -1,103 +1,103 @@
 1.26827e+10
-11
+13
 1
-49.1923,10.2308,0,0,34.8462
+49.25,10.3214,0,0,35.5714
 39.5,12.5,99999,0,45
-23,10.1667,0,0,21.6667
-19.25,9.375,0,0,22.375
-18,5,0,0,22.5
+19.5,6.75,0,0,19.5
+15,6.5,0,0,22.5
+18.75,6.5,0,0,20
 36.5,10,3946,0,32.5
 46.5,9,0,1914,40
 56,10,7420.5,0,47.5
 40.75,12.375,61495.9,0,41.125
-19,9.33333,0,0,17.3333
+11.75,4.5,0,0,20
 42.8333,10.5,5195.5,0,46.8333
 71.5,9.5,0,3400,42.5
-51,9.5,457,0,30
-18,9,0,0,20
-18.5,6.5,0,0,18.75
+50,7.5,457,0,32.5
+22,6,0,0,30
+27.25,7.75,0,0,30
 34.5,9,0,1383.5,42.5
-15.75,4.75,0,0,17.5
-15,6.5,0,0,22.5
+24.75,6,0,0,30
+16.5,4.5,0,0,21.25
 40.375,12,6173.62,0,44
-13.5,5,0,0,20
-44,11.8333,1542.83,0,46.6667
+34,9,0,0,29.5
+43.8333,12.1667,1542.83,0,46.6667
 54.8333,10,0,2165.17,44.1667
-42.1667,9.66667,0,746.25,41.1667
-25.5,9.5,1087,0,32.5
+42.75,9.16667,0,746.25,41.1667
+55.25,10,0,933.75,36.75
 50,14,12293,0,50
-28.1667,10.1667,0,0,54.7667
-29.5,11.75,1203.5,0,32.5
-54,9.75,3340.75,0,38.25
+27.9667,10.1333,0,0,54.6
+38.25,10.875,1159.75,0,25.125
+53.75,12,3340.75,0,40.75
 53.5,9,4402,0,40
 31,11,6813.5,0,40
-23,5.75,0,0,20
-39,8.2,1941.3,0,42.6
+42.5,10,0,0,9
+38.6,7.3,1941.3,0,42.6
 42.25,7.5,5841,0,48
-29,8,0,0,16.25
-25.2,10.4,0,0,26.2
-23.2917,10.75,0,0,43
+20.3333,9.33333,0,0,29.3333
+25.1667,10.5,0,0,25
+9.5,5.25,0,0,7.75
 34.5,11.5,14485.5,0,45
-43.1667,12.6667,2193,0,41.6667
-37.24,10.06,0,0,44.96
+42.5,10.5,2193,0,41.6667
+37.7292,9.91667,0,0,44.9375
 42,12.25,11163.2,0,46
 47,10.5,0,2043.25,41
 26.5,9,4176.5,0,35
-46,8.25,2589,0,41.25
-50.4545,10.8636,0,0,56.1364
+45.5,8.5,2589,0,41.25
+50.65,11,0,0,56.65
 34,13.5,4947.5,0,35.5
 70,6,898.5,0,40
-20,8.25,0,0,29.5
+22.5,5.75,3093.75,0,18.75
 46.875,9.875,0,0,88.25
-27.44,9.52,0,0,39.68
-24,4.25,2568.75,0,22
-63.6,10.3,0,0,16.6
-28.5,9.5,2325,0,32.5
-53.5,12.75,0,933.75,41.25
+27.14,9.18,0,0,39.52
+42.5,9.25,0,852.25,46.25
+62.25,11,0,0,17
+30.5,9.5,2325,0,35
+24.8333,9.66667,0,867.167,28.6667
 44.8333,11.6667,0,1821.33,40.8333
-11.25,4.5,0,0,18.75
-47.9231,9.73077,0,0,45.2308
+17.5,7,0,0,30
+47.8333,9.63333,0,0,45.8667
 41,14,0,1686,53.5
 64.8636,8.27273,0,0,39.9545
 39.25,10.75,0,1738.75,53.75
-26.6667,7.83333,0,0,45.8333
-28,12.5,0,0,9
-32.7027,9.59459,0,0,38.9189
-18.3,8.2,0,0,12.1
-23.8333,12,0,0,16.5
-27.5,3.25,0,0,20
-38,12.4762,0,0,39.3571
-20.8333,5.66667,0,0,40.5
-42.5,10,0,0,9
-67.5,8.8,0,0,5.2
+29.25,4,0,0,32
+27.6,11.6,0,0,13.5
+32.6892,9.63514,0,0,39.2432
+26.25,6.25,0,862.75,33.75
+22.5833,10.3333,0,0,21.25
+39.25,5.25,0,0,12.5
+37.675,12.775,0,0,39.725
+17.75,5.5,0,0,27
+22.75,6.5,831.25,0,22.5
+65.6667,8.83333,0,0,5.5
 56.0278,9.30556,0,0,39.4167
-17,6.5,0,0,27.5
-24.8333,9.16667,0,867.167,29.1667
+27,4.5,572.5,0,7
+17.25,6.75,0,0,14.75
 25.5,11,0,1869.5,41
-45,9.1,0,0,22.6
-20.7667,9.36667,0,0,39.4
-37.375,11.0833,0,0,54.8333
+45.9,8.8,0,0,22.6
+22.5,9.72,0,0,41.16
+37,11.0833,0,0,54.9167
 82,8.5,0,0,48
 33.7778,11.5,0,0,68.4444
-44.25,10.4167,0,0,39.7083
+44.375,11.15,0,0,39.85
 58.2273,10.8636,0,0,46.1364
 39.5,11,0,1944.5,55
-24.1304,11.3913,0,0,35.5217
-28.825,12.225,0,0,43.4
-38.8611,9.16667,0,0,34
+24.9,12.1333,0,0,35.7
+28.375,11.875,0,0,43.7292
+39.2619,9.09524,0,0,34.8571
 59.5,7.33333,0,0,63.1667
-21.5,7,1096.5,0,20
+15,7.5,0,0,16.25
 51.5,13.5,15024,0,40
-12.25,4.5,0,0,22
-59.5,9,1145,0,7
+17,6,0,0,5
+9,4,0,0,4
 80.1667,8.83333,0,0,25.6667
-15.75,5.75,0,0,21.25
-19.625,8.75,0,0,6.25
-32.25,12,0,0,20.75
-34,8,0,0,27.1667
-63.6,7.5,0,0,30.2
-22,4,0,0,44.75
+19.6,10.1,0,0,16.7
+19.5833,9.08333,0,0,8.83333
+18.25,8.875,0,0,23
+29,12,0,0,31.8333
+65.1,7.2,0,0,29.2
+19.7,9.25,0,0,35.7
 30.6667,12.5,0,0,94.5
-16,4.75,0,0,26.25
-15.5,5.25,0,0,19
-42.75,7.25,0,852.25,46.25
+18.25,4,977,0,21.5
+32,7.5,0,0,17
+19.25,6.75,0,0,20

tests/datasets/adult-balanced_50.output

@@ -1,53 +1,53 @@
-1.26829e+10
-20
+1.26828e+10
+25
 1
-68.4667,9.2,0,0,16.1333
+68.8333,9.03333,0,0,16.6333
 39.5,12.5,99999,0,45
-24.4,10.55,0,0,21.9
+20.2857,10.0714,0,0,16.5714
 52,14,15024,0,50
 40.75,12.375,61495.9,0,41.125
 36.5,10,3946,0,32.5
 39.1,11,0,1849.5,40.7
-20.5,10.4286,0,0,15.5714
-18.9167,8.41667,0,0,7.58333
-25.471,10.5145,0,0,41.2681
+38.4286,10.3286,0,0,35.6714
+17.25,6.75,0,0,14.75
+21.7679,9.80357,0,0,35.1964
 42.1667,12.3333,11163.2,0,44.3333
 71.5,9.5,0,3400,42.5
 40.375,12,6173.62,0,44
-16.5,5,0,0,21.5
-56,9.5,457,0,35
+27.8772,10.5702,0,0,39.6754
+52.5,10.5,457,0,35
 34.5,11.5,14485.5,0,45
-43.2,9.3,1941.3,0,45.3
-17.25,6.75,0,0,14.75
+39.4,8.4,1941.3,0,42.6
+27.6,9.65,0,0,29.35
 42.8333,10.5,5195.5,0,46.8333
 56,10,7420.5,0,47.5
-16.5,5.5,0,0,20
+18.9167,8.41667,0,0,7.58333
 48.5,13,12293,0,50
-46.8333,11.3333,2193,0,41.6667
-57.25,7.25,3340.75,0,49.75
-44.5,9.16667,1184,0,20.8333
-39,9.65385,0,819.308,37.7308
+48,11.5,2193,0,40.6667
+55.5,7.5,3340.75,0,52
+38.125,10.125,1159.75,0,25.75
+46.1667,8.5,0,746.25,41.1667
 54.8333,10,0,2165.17,44.1667
-33.373,10.4603,0,0,41.0952
+35.0278,10.5,0,0,42.3704
 53.5,9,4402,0,40
-29,11.5,2325,0,36.5
+30.5,9.5,2325,0,37.5
 43.5,11.8333,0,1721.17,53.6667
 31,11,6813.5,0,40
-47,12,1542.83,0,50
-21.3043,9.34783,0,0,34.4783
-42.0357,10.4214,0,0,42.1786
-47.25,9,2589,0,41.25
+48,11.3333,1542.83,0,50.6667
+18.25,5.5,0,0,18
+35.3333,11.2879,0,0,58
+47.25,9,2589,0,36.875
 42.25,7.5,5841,0,48
 44.5,10.6667,0,2010.33,45.6667
-48.1731,10.0577,0,0,32.6346
-25.5,11,1087,0,35
-18.7857,8.92857,0,0,25
-33.8333,10,0,0,14
-31.6316,9.02632,0,0,32.4474
+45.4151,10.4057,0,0,42.283
+36.6429,8.07143,0,881.929,39.4286
+48,9.2,0,0,23.25
+28.625,11.0625,0,0,14.1875
+20.4545,9.40909,0,0,24.0909
 34,13.5,4947.5,0,35.5
 26.5,9,4176.5,0,35
-33.6618,10.9559,0,0,58.2206
-58.6176,9.38235,0,0,43.6275
+24.1552,9.98276,0,0,45.1379
+58.97,9.36,0,0,42.98
 39,10.5,0,1383.5,45
-40.1111,10.5,0,0,87.3333
+38.5,10.6111,0,0,87.6111
 79.5,8.5,898.5,0,40