From 0d07376d81a85d2a23c4dce276e307051280f55a Mon Sep 17 00:00:00 2001
From: Boris Stefanovic <owldev@bluewin.ch>
Date: Fri, 10 Jun 2022 09:58:18 +0200
Subject: [PATCH] DEBUG: removed cause of infinite loop in kmeans_init_clusters
---
Makefile | 6 +--
src/cluster.c | 71 ++++++++++++-----------------
src/io.c | 6 ++-
src/kmeans.c | 124 ++++++++++++++++++--------------------------------
src/main.c | 47 +++++++------------
src/vector.c | 23 ++++++++++
src/vector.h | 5 ++
test/data.txt | 7 +++
8 files changed, 131 insertions(+), 158 deletions(-)
create mode 100644 test/data.txt
diff --git a/Makefile b/Makefile
index 5163eac..62d5a46 100644
--- a/Makefile
+++ b/Makefile
@@ -14,7 +14,7 @@ OBJ := $(patsubst ${SRC_DIR}/%.c,${BUILD_DIR}/%.o,${SRC})
DEBUG_BUILD_DIR := ${BUILD_ROOT}/debug
DEBUG_TARGET := ${DEBUG_BUILD_DIR}/debug
DEBUG_CFLAGS := ${CFLAGS} -fsanitize=address -fsanitize=leak -g -DDEBUG
-DEBUG_LDEXTRA :=
+DEBUG_LDEXTRA := ${LDEXTRA}
DEBUG_LDFLAGS := ${DEBUG_CFLAGS} ${DEBUG_LDEXTRA}
DEBUG_OBJ := $(patsubst ${SRC_DIR}/%.c,${DEBUG_BUILD_DIR}/%.o,${SRC})
@@ -22,7 +22,7 @@ DEBUG_OBJ := $(patsubst ${SRC_DIR}/%.c,${DEBUG_BUILD_DIR}/%.o,${SRC})
# TARGETS
-all: ${TARGET} ${TARGET_DEBUG}
+all: ${TARGET} ${DEBUG_TARGET}
${TARGET}: ${OBJ}
${CC} ${LDFLAGS} -o $@ $^
@@ -53,7 +53,7 @@ clean:
rm -rf ${BUILD_ROOT}
debug: ${DEBUG_TARGET}
- ./$<
+ ./$< -i test/data.txt
exec: ${TARGET}
./$<
diff --git a/src/cluster.c b/src/cluster.c
index 504c930..446062b 100644
--- a/src/cluster.c
+++ b/src/cluster.c
@@ -2,6 +2,10 @@
#include <stdbool.h>
#include "vector.h"
+#ifdef DEBUG
+#include <assert.h>
+#endif
+
cluster_int_t* cluster_create_int(vector_int_t* center) {
cluster_int_t* cluster = malloc(sizeof(cluster_int_t));
@@ -47,30 +51,8 @@ void cluster_add_point_fpt(cluster_fpt_t* cluster, vector_fpt_t* point) {
bool cluster_update_center_int(cluster_int_t* cluster) {
- // save old center
- vector_int_t* old_center = cluster->center;
- // create new center
- list_points_node_int_t* node = cluster->points->head;
- cluster->center = vector_create_int(node->point->dim);
- // sum all values in center
- while (node != NULL) {
- vector_add_inplace_int(cluster->center, *(node->point));
- node = node->next;
- }
- // divide by number of points
- vector_div_inplace_int(cluster->center, (int_t) cluster->points->size);
- // check whether center has changed
- bool changed = false;
- for (size_t p = 0; p < cluster->center->dim; ++p) {
- if (cluster->center->data[p] != old_center->data[p]) {
- changed = true;
- break;
- }
- }
- // destroy old center
- vector_destroy_int(old_center);
- // return true if center has changed
- return changed;
+ //TODO
+ return true;
}
bool cluster_update_center_fpt(cluster_fpt_t* cluster) {
@@ -78,26 +60,31 @@ bool cluster_update_center_fpt(cluster_fpt_t* cluster) {
vector_fpt_t* old_center = cluster->center;
// create new center
list_points_node_fpt_t* node = cluster->points->head;
- cluster->center = vector_create_fpt(node->point->dim);
- // sum all values in center
- while (node != NULL) {
- vector_add_inplace_fpt(cluster->center, *(node->point));
- node = node->next;
- }
- // divide by number of points
- vector_div_inplace_fpt(cluster->center, (fpt_t) cluster->points->size);
- // check whether center has changed
- bool changed = false;
- for (size_t p = 0; p < cluster->center->dim; ++p) {
- if (cluster->center->data[p] != old_center->data[p]) {
- changed = true;
- break;
+ // if cluster is empty
+ if (NULL == node) {
+ return false; // center has not been changed
+ } else {
+ cluster->center = vector_create_fpt(node->point->dim);
+ // sum all values in center
+ while (node != NULL) {
+ vector_add_inplace_fpt(cluster->center, *(node->point));
+ node = node->next;
+ }
+ // divide by number of points
+ vector_div_inplace_fpt(cluster->center, (fpt_t) cluster->points->size);
+ // check whether center has changed
+ bool changed = false;
+ for (size_t p = 0; p < cluster->center->dim; ++p) {
+ if (cluster->center->data[p] != old_center->data[p]) {
+ changed = true;
+ break;
+ }
}
+ // destroy old center
+ vector_destroy_fpt(old_center);
+ // return true if center has changed
+ return changed;
}
- // destroy old center
- vector_destroy_fpt(old_center);
- // return true if center has changed
- return changed;
}
diff --git a/src/io.c b/src/io.c
index a3674bb..cc7824d 100644
--- a/src/io.c
+++ b/src/io.c
@@ -56,9 +56,10 @@ list_points_int_t* io_get_vector_list_int(FILE* ifile, const size_t dim) {
if (len != 0) {
vector_int_t* vector = io_line_to_vector_int(line, dim);
list_points_append_int(list, vector);
- free(line);
+ //free(line);
}
}
+ free(line);
return list;
}
@@ -70,9 +71,10 @@ list_points_fpt_t* io_get_vector_list_fpt(FILE* ifile, const size_t dim) {
if (len != 0) {
vector_fpt_t* vector = io_line_to_vector_fpt(line, dim);
list_points_append_fpt(list, vector);
- free(line);
+ //free(line);
}
}
+ free(line);
return list;
}
diff --git a/src/kmeans.c b/src/kmeans.c
index 1179541..cb572cd 100644
--- a/src/kmeans.c
+++ b/src/kmeans.c
@@ -1,55 +1,43 @@
#include "kmeans.h"
+#include "vector.h"
+
+#ifdef DEBUG
+#include <assert.h>
+#include "io.h"
+#endif
+
+#define EPSILON 0.001
+
+
+fpt_t abs_fpt(const fpt_t x) {
+ return x >= 0 ? x : -x;
+}
cluster_int_t** kmeans_init_clusters_int(const vector_int_t** points, const size_t point_count, const size_t nclusters) {
- // check args and init
- if (NULL == points || point_count < 2 || nclusters < 2) return NULL;
- cluster_int_t** clusters = calloc(nclusters, sizeof(vector_int_t*));
- if (NULL == clusters) return NULL;
- // determine range in which we are working
- vector_int_t* min = vector_copy_int(points[0]);
- vector_int_t* max = vector_copy_int(points[0]);
- for (size_t i = 1; i < point_count; ++i) {
- for (size_t p = 0; p < max->dim; ++p) {
- const int_t value = points[i]->data[p];
- if (value < min->data[p]) min->data[p] = value;
- if (value > max->data[p]) max->data[p] = value;
- }
- }
- // until we have enough centers
- for (size_t i = 0; i < nclusters; ++i) {
- vector_int_t* center = vector_create_int(max->dim);
- bool valid = false;
- while (!valid) {
- // initialise center values randomly, within the "polygon" of our set of points
- for (size_t p = 0; p < center->dim; ++p) {
- center->data[p] = rand_int_range(min->data[p], max->data[p] + 1);
- }
- // check center is not already in clusters, although probability is extremely low...
- for (size_t k = 0; k < i; ++k) {
- vector_int_t* kth_center = clusters[k]->center;
- bool neq = false;
- for (size_t p = 0; p < center->dim; ++p) {
- if (center->data[p] != kth_center->data[p]) {
- neq = true;
- break;
- }
- }
- if (neq) {
- valid = true;
- }
- }
+ //TODO
+ return NULL;
+}
+
+
+bool is_vector_in_centers_fpt(const vector_fpt_t* center, const cluster_fpt_t** clusters, const size_t i) {
+ for (size_t k = 0; k < i; ++k) {
+ if (vector_equals_fpt(clusters[k]->center, center)) {
+ return true;
}
- clusters[i]->center = center;
}
- return clusters;
+ return false;
}
+
cluster_fpt_t** kmeans_init_clusters_fpt(const vector_fpt_t** points, const size_t point_count, const size_t nclusters) {
// check args and init
if (NULL == points || point_count < 2 || nclusters < 2) return NULL;
cluster_fpt_t** clusters = calloc(nclusters, sizeof(vector_fpt_t*));
if (NULL == clusters) return NULL;
+ for (size_t k = 0; k < nclusters; ++k) {
+ clusters[k] = cluster_create_fpt(NULL);
+ }
// determine range in which we are working
vector_fpt_t* min = vector_copy_fpt(points[0]);
vector_fpt_t* max = vector_copy_fpt(points[0]);
@@ -65,26 +53,14 @@ cluster_fpt_t** kmeans_init_clusters_fpt(const vector_fpt_t** points, const size
vector_fpt_t* center = vector_create_fpt(max->dim);
bool valid = false;
while (!valid) {
- // initialise center values randomly, within the "polygon" of our set of points
+ // initialise center values randomly, within the "multidimensional rectangle" of our set of points
for (size_t p = 0; p < center->dim; ++p) {
center->data[p] = rand_double_range(min->data[p], max->data[p]);
}
// check center is not already in clusters, although probability is extremely low...
- for (size_t k = 0; k < i; ++k) {
- vector_fpt_t* kth_center = clusters[k]->center;
- bool neq = false;
- for (size_t p = 0; p < center->dim; ++p) {
- if (center->data[p] != kth_center->data[p]) {
- neq = true;
- break;
- }
- }
- if (neq) {
- valid = true;
- }
- }
+ valid = !is_vector_in_centers_fpt(center, (const cluster_fpt_t**) clusters, i);
}
- clusters[i]->center = center;
+ clusters[i] = cluster_create_fpt(center);
}
return clusters;
}
@@ -92,41 +68,24 @@ cluster_fpt_t** kmeans_init_clusters_fpt(const vector_fpt_t** points, const size
void kmeans_int(vector_int_t** points, const size_t point_count, cluster_int_t** clusters, const size_t nb_clusters,
fpt_t (* distance_function)(const vector_int_t*, const vector_int_t*)) {
- bool changed = true;
- while (changed) {
- changed = false;
- for (size_t i = 0; i < point_count; ++i) {
- vector_int_t* point = points[i];
- // find closest cluster and add point to it
- cluster_int_t* cmin = clusters[0];
- int_t dmin = distance_function(point, cmin->center);
- for (size_t k = 1; k < nb_clusters; ++k) {
- cluster_int_t* current_cluster = clusters[k];
- fpt_t dist = distance_function(point, current_cluster->center);
- if (dist < dmin) {
- cmin = current_cluster;
- dmin = dist;
- }
- }
- cluster_add_point_int(cmin, point);
- // update all cluster centers
- for (size_t k = 0; k < nb_clusters; ++k) {
- if (cluster_update_center_int(clusters[k])) {
- changed = true;
- }
- }
- }
- }
+ //TODO
}
+
void kmeans_fpt(vector_fpt_t** points, const size_t point_count, cluster_fpt_t** clusters, const size_t nb_clusters,
fpt_t (* distance_function)(const vector_fpt_t*, const vector_fpt_t*)) {
bool changed = true;
while (changed) {
+ // reset condition
changed = false;
+ // empty all clusters, keeping only their centers (virtual)
+ for (size_t k = 0; k < nb_clusters; ++k) {
+ cluster_reset_fpt(clusters[k]);
+ }
+ // for each point
for (size_t i = 0; i < point_count; ++i) {
vector_fpt_t* point = points[i];
- // find closest cluster and add point to it
+ // find closest cluster
cluster_fpt_t* cmin = clusters[0];
fpt_t dmin = distance_function(point, cmin->center);
for (size_t k = 1; k < nb_clusters; ++k) {
@@ -137,9 +96,14 @@ void kmeans_fpt(vector_fpt_t** points, const size_t point_count, cluster_fpt_t**
dmin = dist;
}
}
+ // add point to closest cluster
cluster_add_point_fpt(cmin, point);
// update all cluster centers
for (size_t k = 0; k < nb_clusters; ++k) {
+#ifdef DEBUG
+ assert(clusters[k] != NULL);
+ assert(clusters[k]->points != NULL);
+#endif
if (cluster_update_center_fpt(clusters[k])) {
changed = true;
}
diff --git a/src/main.c b/src/main.c
index c12fe2c..5df68ee 100644
--- a/src/main.c
+++ b/src/main.c
@@ -9,6 +9,10 @@
#include "linkedlist.h"
#include "vector.h"
+#ifdef DEBUG
+#include <assert.h>
+#endif
+
enum DistanceFunctionType {
EUCLID = 0, MANHATTAN = 1, CHEBYSHEV = 2
@@ -60,7 +64,7 @@ void parse_args(int argc, char** argv, char** ipath, char** opath, enum Distance
else if (strcmp(optarg, "int") == 0) *type = INT;
break;
case '?':
- //TODO: perhaps add an "unknown option" message on stderr
+ fprintf(stderr, "UNKNOWN OPTION : %c", opt);
break;
default:
// https://www.gnu.org/software/libc/manual/html_node/Example-of-Getopt.html
@@ -71,36 +75,8 @@ void parse_args(int argc, char** argv, char** ipath, char** opath, enum Distance
int main_int(const char* ipath, const char* opath, const enum DistanceFunctionType dist_func_type) {
- // READ
- FILE* ifile = ipath != NULL ? fopen(ipath, "r") : stdin;
- const size_t dim = io_read_int(ifile);
- const size_t nb_clusters = io_read_int(ifile);
- if (0 == dim) {
- printf("DIMENSION MUST BE STRICTLY POSITIVE !\n");
- fclose(ifile);
- return EXIT_FAILURE;
- }
- if (0 == nb_clusters) {
- printf("NUMBER OF CLUSTERS MUST BE STRICTLY POSITIVE !\n");
- fclose(ifile);
- return EXIT_FAILURE;
- }
- list_points_int_t* list = io_get_vector_list_int(ifile, dim);
- fclose(ifile);
- ifile = NULL;
- const size_t point_count = list->size;
- vector_int_t** points = list_points_to_array_int(list);
- list_points_destroy_int(list, false);
- list = NULL;
- // ALGORITHM
- cluster_int_t** clusters = kmeans_init_clusters_int((const vector_int_t**) points, point_count, nb_clusters);
- kmeans_int(points, point_count, clusters, nb_clusters, DIST_FUNC_INT[dist_func_type]);
- // WRITE
- FILE* ofile = opath != NULL ? fopen(opath, "w") : stdout;
- fprintf(ofile, "%lud\n%lud\n", dim, nb_clusters);
- io_write_clusters_to_file_int(ofile, clusters, point_count);
- fclose(ofile);
- return EXIT_SUCCESS;
+ //TODO
+ return EXIT_FAILURE;
}
int main_fpt(const char* ipath, const char* opath, const enum DistanceFunctionType dist_func_type) {
@@ -126,8 +102,15 @@ int main_fpt(const char* ipath, const char* opath, const enum DistanceFunctionTy
list_points_destroy_fpt(list, false);
list = NULL;
// ALGORITHM
+ printf("INIT: ... ");
cluster_fpt_t** clusters = kmeans_init_clusters_fpt((const vector_fpt_t**) points, point_count, nb_clusters);
+ printf("DONE\n");
+#ifdef DEBUG
+ for(size_t i = 0; i < nb_clusters; ++i) assert(clusters[i] !=NULL);
+#endif
+ printf("STARTING KMEANS ALGORITHM: ...\n");
kmeans_fpt(points, point_count, clusters, nb_clusters, DIST_FUNC_FPT[dist_func_type]);
+ printf("KMEANS DONE !\n");
// WRITE
FILE* ofile = opath != NULL ? fopen(opath, "w") : stdout;
fprintf(ofile, "%lud\n%lud\n", dim, nb_clusters);
@@ -146,8 +129,10 @@ int main(int argc, char** argv) {
parse_args(argc, argv, &ipath, &opath, &disttype, &datatype);
switch (datatype) {
case FLOAT:
+ printf("FLOAT\n");
return main_fpt(ipath, opath, disttype);
case INT:
+ printf("INT\n");
return main_int(ipath, opath, disttype);
default:
abort();
diff --git a/src/vector.c b/src/vector.c
index b70c9f1..6d31b50 100644
--- a/src/vector.c
+++ b/src/vector.c
@@ -111,3 +111,26 @@ void vector_div_inplace_fpt(vector_fpt_t* v, const fpt_t a) {
if (NULL == v) return;
for (size_t i = 0; i < v->dim; ++i) v->data[i] /= a;
}
+
+
+void vector_print_int(const vector_int_t* v) {
+ if (NULL == v) printf("NULL");
+ else {
+ printf("%ld", v->data[0]);
+ for (size_t p = 1; p < v->dim; ++p) {
+ printf(" , %ld", v->data[p]);
+ }
+ printf("\n");
+ }
+}
+
+void vector_print_fpt(const vector_fpt_t* v) {
+ if (NULL == v) printf("NULL");
+ else {
+ printf("%lf", v->data[0]);
+ for (size_t p = 1; p < v->dim; ++p) {
+ printf(" , %lf", v->data[p]);
+ }
+ printf("\n");
+ }
+}
diff --git a/src/vector.h b/src/vector.h
index 155fb78..7eed5ce 100644
--- a/src/vector.h
+++ b/src/vector.h
@@ -54,4 +54,9 @@ void vector_div_inplace_int(vector_int_t* v, const int_t a);
void vector_div_inplace_fpt(vector_fpt_t* v, const fpt_t a);
+void vector_print_int(const vector_int_t* v);
+
+void vector_print_fpt(const vector_fpt_t* v);
+
+
#endif //PROG_KMEANS_VECTOR_H
diff --git a/test/data.txt b/test/data.txt
new file mode 100644
index 0000000..ffea229
--- /dev/null
+++ b/test/data.txt
@@ -0,0 +1,7 @@
+3
+3
+13, 6, 7
+100.5, 78.32, 1012.34
+-1, -1, -1
+14.2, 5.7, 7.56
+99, 79, 1011
--
GitLab