Versions of algorithms implementing tighter LB update and other improvements

fast_kmeans/Makefile

@@ -7,13 +7,15 @@ CPPFLAGS += -g
 CPPFLAGS += -O3
 CPPFLAGS += -Wno-long-long
 
+CPPFLAGS += -std=c++0x
+
 # Verify algorithm correctness while debugging
 #CPPFLAGS += -DDEBUG
 #CPPFLAGS += -DVERIFY_ASSIGNMENTS
 
 # To use pthreads, uncomment both lines below.
 #CPPFLAGS += -DUSE_THREADS
-#LDFLAGS += -lpthread
+#CPPFLAGS += -lpthread
 
 # Monitor internal algorithm effectiveness
 #CPPFLAGS += -DCOUNT_DISTANCES

fast_kmeans/annulus_kmeans_modified.cpp

@@ -0,0 +1,136 @@
+/* Authors: Greg Hamerly and Jonathan Drake and Petr Ryšavý
+ * Feedback: [email protected]
+ * See: http://cs.baylor.edu/~hamerly/software/kmeans.php
+ * Copyright 2015
+ */
+
+#include "annulus_kmeans_modified.h"
+#include "general_functions.h"
+#include <cmath>
+#include <numeric>
+#include <algorithm>
+
+void AnnulusKmeansModified::free() {
+    HamerlyKmeansModified::free();
+
+    delete [] guard;
+    delete [] xNorm;
+    delete [] cOrder;
+    guard = NULL;
+    xNorm = NULL;
+    cOrder = NULL;
+}
+
+int AnnulusKmeansModified::runThread(int threadId, int maxIterations) {
+    int iterations = 0;
+
+    int startNdx = start(threadId);
+    int endNdx = end(threadId);
+
+    // here we need to calculate s & the centroid-centroid distances before the first iteration
+    // the remaining calls to this method are hidden by move_centers
+    update_s(threadId);
+    synchronizeAllThreads();
+
+    while ((iterations < maxIterations) && !converged) {
+        ++iterations;
+
+        if (threadId == 0) {
+            sort_means_by_norm();
+        }
+        synchronizeAllThreads();
+
+        for (int i = startNdx; i < endNdx; ++i) {
+            unsigned short closest = assignment[i];
+
+            double upper_comparison_bound = std::max(s[closest], lower[i]);
+
+            if (upper[i] <= upper_comparison_bound) {
+                continue;
+            }
+
+            double u2 = pointCenterDist2(i, closest);
+            upper[i] = sqrt(u2);
+
+            if (upper[i] <= upper_comparison_bound) {
+                continue;
+            }
+
+            double l2 = pointCenterDist2(i, guard[i]);
+            lower[i] = sqrt(l2);
+
+            double beta = std::max(lower[i], upper[i]);
+
+            std::pair<double, int>* begin = std::lower_bound(cOrder, cOrder + k, std::make_pair(xNorm[i] - beta, k));
+            std::pair<double, int>* end = std::lower_bound(begin, cOrder + k, std::make_pair(xNorm[i] + beta, k));
+
+            for (std::pair<double, int>* jp = begin; jp != end; ++jp) {
+                if (jp->second == closest) continue;
+
+                double dist2 = pointCenterDist2(i, jp->second);
+                if (dist2 <= u2) {
+                    if (dist2 == u2) {
+                        if (jp->second < closest) closest = jp->second;
+                    } else {
+                        l2 = u2;
+                        u2 = dist2;
+                        guard[i] = closest;
+                        closest = jp->second;
+                    }
+                } else if (dist2 < l2) {
+                    l2 = dist2;
+                    guard[i] = jp->second;
+                }
+            }
+
+            lower[i] = sqrt(l2);
+
+            if (assignment[i] != closest) {
+                upper[i] = sqrt(u2);
+                changeAssignment(i, closest, threadId);
+            }
+        }
+
+        verifyAssignment(iterations, startNdx, endNdx);
+
+        synchronizeAllThreads();
+        move_centers(threadId);
+
+        synchronizeAllThreads();
+        if (!converged) {
+            update_bounds(startNdx, endNdx);
+        }
+
+        synchronizeAllThreads();
+    }
+
+    return iterations;
+}
+
+void AnnulusKmeansModified::initialize(Dataset const *aX, unsigned short aK, unsigned short *initialAssignment, int aNumThreads) {
+    HamerlyKmeansModified::initialize(aX, aK, initialAssignment, aNumThreads);
+
+    guard = new unsigned short[n];
+    xNorm = new double[n];
+    cOrder = new std::pair<double, int>[k];
+
+    for (int i = 0; i < k; ++i) {
+        cOrder[i].first = 0.0;
+        cOrder[i].second = i;
+    }
+
+    std::fill(guard, guard + n, 1);
+    for (int i = 0; i < n; ++i) {
+        xNorm[i] = sqrt(pointPointInnerProduct(i, i));
+    }
+}
+
+void AnnulusKmeansModified::sort_means_by_norm() {
+    for (int c1 = 0; c1 < k; ++c1) {
+        cOrder[c1].first = sqrt(centerCenterInnerProduct(c1, c1));
+        cOrder[c1].second = c1;
+    }
+    std::sort(cOrder, cOrder + k);
+}
+
+

fast_kmeans/annulus_kmeans_modified.h

@@ -0,0 +1,39 @@
+#ifndef ANNULUS_KMEANS_MODIFIED_H
+#define ANNULUS_KMEANS_MODIFIED_H
+
+/* Authors: Greg Hamerly and Jonathan Drake and Petr Ryšavý
+ * Feedback: [email protected]
+ * See: http://cs.baylor.edu/~hamerly/software/kmeans.php
+ * Copyright 2015
+ *
+ * This version of Annulus algorithm implements the tighter update. Most of the
+ * code is copied from the default version of Annulus algorithm, only the parent
+ * class is HamerlyKmeansModified.
+ */
+
+#include "hamerly_kmeans_modified.h"
+#include <utility>
+
+class AnnulusKmeansModified;
+
+class AnnulusKmeansModified : public HamerlyKmeansModified {
+    public:
+        AnnulusKmeansModified() : xNorm(NULL), cOrder(NULL), guard(NULL) {}
+        virtual ~AnnulusKmeansModified() { free(); }
+        virtual void free();
+        virtual void initialize(Dataset const *aX, unsigned short aK, unsigned short *initialAssignment, int aNumThreads);
+        virtual std::string getName() const { return "annulusmodified"; }
+
+    protected:
+        virtual int runThread(int threadId, int maxIterations);
+        void sort_means_by_norm();
+
+        double *xNorm;
+
+        std::pair<double, int> *cOrder;
+
+        unsigned short *guard;
+};
+
+#endif
+

fast_kmeans/dataset.h

@@ -37,10 +37,9 @@ class Dataset {
         // destroys the dataset safely
         ~Dataset() {
             n = d = nd = 0; 
-            double *dp = data, *sdsp = sumDataSquared;
+			delete [] data;
+			delete [] sumDataSquared;
             data = sumDataSquared = NULL;
-            delete [] dp;
-            delete [] sdsp;
         }
 
         // operator= is the standard deep-copy assignment operator, which