/**
* This application groups 'num_points' row-vectors (which are randomly
* generated) into 'num_means' clusters through an iterative algorithm - the
* k-means algorith
*/
int main(int argc, char **argv)
{
int **points;
int **means;
int *clusters;
int i;
parse_args(argc, argv);
points = (int **)malloc(sizeof(int *) * num_points);
for (i=0; i<num_points; i++)
{
points[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating points\n");
generate_points(points, num_points);
means = (int **)malloc(sizeof(int *) * num_means);
for (i=0; i<num_means; i++)
{
means[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating means\n");
generate_points(means, num_means);
clusters = (int *)malloc(sizeof(int) * num_points);
memset(clusters, -1, sizeof(int) * num_points);
modified = true;
dprintf("\n\nStarting iterative algorithm\n");
while (modified)
{
modified = false;
dprintf(".");
find_clusters(points, means, clusters);
calc_means(points, means, clusters);
}
dprintf("\n\nFinal Means:\n");
dump_matrix(means, num_means, dim);
dprintf("Cleaning up\n");
for (i=0; i<num_means; i++) {
free(means[i]);
}
free(means);
for (i=0; i<num_points; i++) {
free(points[i]);
}
free(points);
return 0;
}
int main()
{
// declare input set
point_t inputs[] =
{
{2.1, 4.3},
{3.2618187593136714, 5.248776670511476},
{4.229518090705041, 6.394882252046046},
{4.970144154070997, 7.699287493626686},
{5.4584758171066685, 9.117572439213015},
{5.677883530716649, 10.601439096028392},
{5.620895628364498, 12.10035616477686},
{5.289452764715364, 13.563279822435065},
{4.694841828968918, 14.940391958424991},
{3.857311583384793, 16.18479667061946},
{2.8053831159695455, 17.25411724904363},
{1.574878589031866, 18.11193926390795},
{0.20770135840910497, 18.729050615340313},
{-1.2495909950582316, 19.08443631641118},
{-2.7473721281023433, 19.165994133308747},
{-4.234636900627881, 18.97094671239639},
{-5.66073829458397, 18.505936159659683},
{-6.9771121374195335, 17.786797851753388},
{-8.138930896733212, 16.83802118124192},
{-9.10663022812459, 15.691915599707357},
{-9.847256291490558, 14.387510358126725}
};
point_t* points = calloc(21, sizeof(point_t));
// print calculated points
generate_points(inputs, points, 0, 21);
}
void setup()
{
PbScreenSetup( SCR_W, SCR_H, SCR_PSM );
build_matrices();
generate_points();
load_texture();
}
int main(int argc, char **argv) {
int i;
parse_args(argc, argv);
// Create the matrix to store the points
matrix = (int **)malloc(sizeof(int *) * num_rows);
for (i=0; i<num_rows; i++)
{
matrix[i] = (int *)malloc(sizeof(int) * num_cols);
}
//Generate random values for all the points in the matrix
generate_points(matrix, num_rows, num_cols);
// Print the points
// dump_points(matrix, num_rows, num_cols);
// Allocate Memory to store the mean and the covariance matrix
mean = (int *)malloc(sizeof(int) * num_rows);
cov = (int **)malloc(sizeof(int *) * num_rows);
for (i=0; i<num_rows; i++)
{
cov[i] = (int *)malloc(sizeof(int) * num_rows);
// printf ("cov[%d] = %p\n", i, cov[i]);
}
// Compute the mean and the covariance
CHECK_ERROR((num_procs = sysconf(_SC_NPROCESSORS_ONLN)) <= 0);
// goes faster with more threads
// maknum_procs = 48;
printf("The number of processors is %d\n", num_procs);
_pthread_mean();
printf ("now spawning cov threads.\n"); fflush (stdout);
_pthread_cov();
dump_points(cov, num_rows, num_rows);
#if 0
for (i=0; i<num_rows; i++)
{
free(cov[i]);
free(matrix[i]);
}
free(mean);
free(cov);
free(matrix);
#endif
return 0;
}
/**
* @brief Initializes the main application class.
*/
ConvexHullApp::ConvexHullApp()
{
/* Do SDL initialization. */
surf = NULL;
running = true;
/* Generate a list of points. */
generate_points();
}
LivePoints(){
mndim = 2;
msize = 10;
mpoints.setZero(mndim,msize);
mlikelihoods.setZero(msize);
generate_points();
}
inline void apply(Point const& penultimate_point,
Point const& perp_left_point,
Point const& ultimate_point,
Point const& ,
buffer_side_selector side,
DistanceStrategy const& distance,
RangeOut& range_out) const
{
typedef typename coordinate_type<Point>::type coordinate_type;
typedef typename geometry::select_most_precise
<
coordinate_type,
double
>::type promoted_type;
promoted_type const alpha = calculate_angle<promoted_type>(perp_left_point, ultimate_point);
promoted_type const dist_left = distance.apply(penultimate_point, ultimate_point, buffer_side_left);
promoted_type const dist_right = distance.apply(penultimate_point, ultimate_point, buffer_side_right);
if (geometry::math::equals(dist_left, dist_right))
{
generate_points(ultimate_point, alpha, dist_left, range_out);
}
else
{
promoted_type const two = 2.0;
promoted_type dist_half_diff = (dist_left - dist_right) / two;
if (side == buffer_side_right)
{
dist_half_diff = -dist_half_diff;
}
Point shifted_point;
set<0>(shifted_point, get<0>(ultimate_point) + dist_half_diff * cos(alpha));
set<1>(shifted_point, get<1>(ultimate_point) + dist_half_diff * sin(alpha));
generate_points(shifted_point, alpha, (dist_left + dist_right) / two, range_out);
}
}
int main(int argc, char **argv) {
int i;
parse_args(argc, argv);
// Create the matrix to store the points
matrix = (int **)malloc(sizeof(int *) * num_rows);
for (i=0; i<num_rows; i++)
{
matrix[i] = (int *)malloc(sizeof(int) * num_cols);
}
//Generate random values for all the points in the matrix
generate_points(matrix, num_rows, num_cols);
// Print the points
#ifndef FAULTINJECTION
dump_points(matrix, num_rows, num_cols);
#endif
// Allocate Memory to store the mean and the covariance matrix
mean = (int *)malloc(sizeof(int) * num_rows);
cov = (int **)malloc(sizeof(int *) * num_rows);
for (i=0; i<num_rows; i++)
{
cov[i] = (int *)malloc(sizeof(int) * num_rows);
}
// Compute the mean and the covariance
pthread_mean();
pthread_cov();
dump_points(cov, num_rows, num_rows);
for (i=0; i<num_rows; i++)
{
free(cov[i]);
free(matrix[i]);
}
free(mean);
free(cov);
free(matrix);
return 0;
}
int main(int argc, char **argv)
{
int num_procs, curr_point;
int i;
pthread_t *pid;
pthread_attr_t attr;
thread_arg *arg;
int num_per_thread, excess;
parse_args(argc, argv);
points = (int **)malloc(sizeof(int *) * num_points);
for (i=0; i<num_points; i++)
{
points[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating points\n");
generate_points(points, num_points);
means = (int **)malloc(sizeof(int *) * num_means);
for (i=0; i<num_means; i++)
{
means[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating means\n");
generate_points(means, num_means);
clusters = (int *)malloc(sizeof(int) * num_points);
memset(clusters, -1, sizeof(int) * num_points);
pthread_attr_init(&attr);
pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM);
CHECK_ERROR((num_procs = sysconf(_SC_NPROCESSORS_ONLN)) <= 0);
CHECK_ERROR( (pid = (pthread_t *)malloc(sizeof(pthread_t) * num_procs)) == NULL);
modified = true;
printf("Starting iterative algorithm\n");
/* Create the threads to process the distances between the various
points and repeat until modified is no longer valid */
int num_threads;
while (modified)
{
num_per_thread = num_points / num_procs;
excess = num_points % num_procs;
modified = false;
dprintf(".");
curr_point = 0;
num_threads = 0;
while (curr_point < num_points) {
CHECK_ERROR((arg = (thread_arg *)malloc(sizeof(thread_arg))) == NULL);
arg->start_idx = curr_point;
arg->num_pts = num_per_thread;
if (excess > 0) {
arg->num_pts++;
excess--;
}
CHECK_ERROR((pthread_create(&(pid[num_threads++]), &attr, find_clusters,
(void *)(arg))) != 0);
curr_point += arg->num_pts;
}
assert (num_threads == num_procs);
for (i = 0; i < num_threads; i++) {
pthread_join(pid[i], NULL);
}
num_per_thread = num_means / num_procs;
excess = num_means % num_procs;
curr_point = 0;
num_threads = 0;
while (curr_point < num_means) {
CHECK_ERROR((arg = (thread_arg *)malloc(sizeof(thread_arg))) == NULL);
arg->start_idx = curr_point;
arg->sum = (int *)malloc(dim * sizeof(int));
arg->num_pts = num_per_thread;
if (excess > 0) {
arg->num_pts++;
excess--;
}
CHECK_ERROR((pthread_create(&(pid[num_threads++]), &attr, calc_means,
(void *)(arg))) != 0);
curr_point += arg->num_pts;
}
assert (num_threads == num_procs);
for (i = 0; i < num_threads; i++) {
pthread_join(pid[i], NULL);
}
}
dprintf("\n\nFinal means:\n");
dump_points(means, num_means);
for (i = 0; i < num_points; i++)
free(points[i]);
free(points);
for (i = 0; i < num_means; i++)
{
free(means[i]);
}
free(means);
free(clusters);
return 0;
}
int main(int argc, char **argv)
{
final_data_t pca_mean_vals;
final_data_t pca_cov_vals;
map_reduce_args_t map_reduce_args;
int i;
struct timeval begin, end;
#ifdef TIMING
unsigned int library_time = 0;
#endif
get_time (&begin);
parse_args(argc, argv);
// Allocate space for the matrix
pca_data.matrix = (int *)malloc(sizeof(int) * num_rows * num_cols);
//Generate random values for all the points in the matrix
generate_points(pca_data.matrix, num_rows, num_cols);
// Print the points
//dump_points(pca_data.matrix, num_rows, num_cols);
/* Create the structure to store the mean value */
pca_data.unit_size = sizeof(int) * num_cols; // size of one row
pca_data.next_start_row = pca_data.next_cov_row = 0;
pca_data.mean = NULL;
CHECK_ERROR (map_reduce_init ());
// Setup scheduler args for computing the mean
memset(&map_reduce_args, 0, sizeof(map_reduce_args_t));
map_reduce_args.task_data = &pca_data;
map_reduce_args.map = pca_mean_map;
map_reduce_args.reduce = NULL; // use identity reduce
map_reduce_args.splitter = pca_mean_splitter;
map_reduce_args.locator = pca_mean_locator;
map_reduce_args.key_cmp = mymeancmp;
map_reduce_args.unit_size = pca_data.unit_size;
map_reduce_args.partition = NULL; // use default
map_reduce_args.result = &pca_mean_vals;
map_reduce_args.data_size = num_rows * num_cols * sizeof(int);
map_reduce_args.L1_cache_size = atoi(GETENV("MR_L1CACHESIZE"));//1024 * 1024 * 16;
map_reduce_args.num_map_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_reduce_threads = atoi(GETENV("MR_NUMTHREADS"));//16;
map_reduce_args.num_merge_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_procs = atoi(GETENV("MR_NUMPROCS"));//16;
map_reduce_args.key_match_factor = (float)atof(GETENV("MR_KEYMATCHFACTOR"));//2;
printf("PCA Mean: Calling MapReduce Scheduler\n");
get_time (&end);
#ifdef TIMING
fprintf (stderr, "initialize: %u\n", time_diff (&end, &begin));
#endif
get_time (&begin);
CHECK_ERROR(map_reduce(&map_reduce_args) < 0);
get_time (&end);
#ifdef TIMING
library_time += time_diff (&end, &begin);
#endif
get_time (&begin);
printf("PCA Mean: MapReduce Completed\n");
assert (pca_mean_vals.length == num_rows);
//dprintf("Mean vector:\n");
pca_data.unit_size = sizeof(int) * num_cols * 2; // size of two rows
pca_data.next_start_row = pca_data.next_cov_row = 0;
pca_data.mean = pca_mean_vals.data; // array of keys and values - the keys have been freed tho
// Setup Scheduler args for computing the covariance
memset(&map_reduce_args, 0, sizeof(map_reduce_args_t));
map_reduce_args.task_data = &pca_data;
map_reduce_args.map = pca_cov_map;
map_reduce_args.reduce = NULL; // use identity reduce
map_reduce_args.splitter = pca_cov_splitter;
map_reduce_args.locator = pca_cov_locator;
map_reduce_args.key_cmp = mycovcmp;
map_reduce_args.unit_size = pca_data.unit_size;
map_reduce_args.partition = NULL; // use default
map_reduce_args.result = &pca_cov_vals;
// data size is number of elements that need to be calculated in a cov matrix
// multiplied by the size of two rows for each element
map_reduce_args.data_size = ((((num_rows * num_rows) - num_rows)/2) + num_rows) * pca_data.unit_size;
map_reduce_args.L1_cache_size = atoi(GETENV("MR_L1CACHESIZE"));//1024 * 1024 * 16;
map_reduce_args.num_map_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_reduce_threads = atoi(GETENV("MR_NUMTHREADS"));//16;
map_reduce_args.num_merge_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_procs = atoi(GETENV("MR_NUMPROCS"));//16;
map_reduce_args.key_match_factor = atoi(GETENV("MR_KEYMATCHFACTOR"));//2;
map_reduce_args.use_one_queue_per_task = true;
printf("PCA Cov: Calling MapReduce Scheduler\n");
get_time (&end);
#ifdef TIMING
fprintf (stderr, "inter library: %u\n", time_diff (&end, &begin));
#endif
get_time (&begin);
CHECK_ERROR(map_reduce(&map_reduce_args) < 0);
get_time (&end);
#ifdef TIMING
library_time += time_diff (&end, &begin);
fprintf (stderr, "library: %u\n", library_time);
#endif
get_time (&begin);
CHECK_ERROR (map_reduce_finalize ());
printf("PCA Cov: MapReduce Completed\n");
assert(pca_cov_vals.length == ((((num_rows * num_rows) - num_rows)/2) + num_rows));
// Free the allocated structures
int cnt = 0;
intptr_t sum = 0;
dprintf("\n\nCovariance sum: ");
for (i = 0; i <pca_cov_vals.length; i++)
{
sum += (intptr_t)(pca_cov_vals.data[i].val);
//dprintf("%5d ", );
cnt++;
if (cnt == num_rows)
{
//dprintf("\n");
num_rows--;
cnt = 0;
}
free(pca_cov_vals.data[i].key);
}
dprintf ("%" PRIdPTR "\n", sum);
free (pca_cov_vals.data);
free (pca_mean_vals.data);
free (pca_data.matrix);
get_time (&end);
#ifdef TIMING
fprintf (stderr, "finalize: %u\n", time_diff (&end, &begin));
#endif
return 0;
}
int main(int argc, char **argv)
{
final_data_t kmeans_vals;
map_reduce_args_t map_reduce_args;
int i;
int *means;
bool first_run;
struct timeval begin, end;
#ifdef TIMING
unsigned int library_time = 0;
unsigned int inter_library_time = 0;
#endif
get_time (&begin);
parse_args(argc, argv);
// get points
kmeans_data.points = (int *)malloc(sizeof(int) * num_points * dim);
generate_points(kmeans_data.points, num_points);
// get means
kmeans_data.means = (keyval_t *)malloc(sizeof(keyval_t) * num_means);
means = malloc(sizeof(int) * dim * num_means);
for (i=0; i<num_means; i++) {
kmeans_data.means[i].val = &means[i * dim];
kmeans_data.means[i].key = malloc(sizeof(void *));
}
generate_means(kmeans_data.means, num_means);
kmeans_data.next_point = 0;
kmeans_data.unit_size = sizeof(int) * dim;
kmeans_data.clusters = (int *)malloc(sizeof(int) * num_points);
memset(kmeans_data.clusters, -1, sizeof(int) * num_points);
modified = true;
CHECK_ERROR (map_reduce_init ());
// Setup map reduce args
memset(&map_reduce_args, 0, sizeof(map_reduce_args_t));
map_reduce_args.task_data = &kmeans_data;
map_reduce_args.map = kmeans_map;
map_reduce_args.reduce = kmeans_reduce;
map_reduce_args.splitter = kmeans_splitter;
map_reduce_args.locator = kmeans_locator;
map_reduce_args.key_cmp = mykeycmp;
map_reduce_args.unit_size = kmeans_data.unit_size;
map_reduce_args.partition = NULL; // use default
map_reduce_args.result = &kmeans_vals;
map_reduce_args.data_size = (num_points + num_means) * dim * sizeof(int);
map_reduce_args.L1_cache_size = atoi(GETENV("MR_L1CACHESIZE"));//1024 * 8;
map_reduce_args.num_map_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_reduce_threads = atoi(GETENV("MR_NUMTHREADS"));//16;
map_reduce_args.num_merge_threads = atoi(GETENV("MR_NUMTHREADS"));//8;
map_reduce_args.num_procs = atoi(GETENV("MR_NUMPROCS"));//16;
map_reduce_args.key_match_factor = (float)atof(GETENV("MR_KEYMATCHFACTOR"));//2;
map_reduce_args.use_one_queue_per_task = true;
printf("KMeans: Calling MapReduce Scheduler\n");
get_time (&end);
#ifdef TIMING
fprintf (stderr, "initialize: %u\n", time_diff (&end, &begin));
#endif
first_run = true;
while (modified == true)
{
modified = false;
kmeans_data.next_point = 0;
//dprintf(".");
get_time (&begin);
CHECK_ERROR (map_reduce (&map_reduce_args) < 0);
get_time (&end);
#ifdef TIMING
library_time += time_diff (&end, &begin);
#endif
get_time (&begin);
for (i = 0; i < kmeans_vals.length; i++)
{
int mean_idx = *((int *)(kmeans_vals.data[i].key));
if (first_run == false)
free(kmeans_data.means[mean_idx].val);
kmeans_data.means[mean_idx] = kmeans_vals.data[i];
}
if (kmeans_vals.length > 0) free(kmeans_vals.data);
get_time (&end);
#ifdef TIMING
inter_library_time += time_diff (&end, &begin);
#endif
first_run = false;
}
#ifdef TIMING
fprintf (stderr, "library: %u\n", library_time);
fprintf (stderr, "inter library: %u\n", inter_library_time);
#endif
get_time (&begin);
CHECK_ERROR (map_reduce_finalize ());
dprintf("\n");
printf("KMeans: MapReduce Completed\n");
dprintf("\n\nFinal means:\n");
dump_means(kmeans_data.means, num_means);
free(kmeans_data.points);
for (i = 0; i < num_means; i++)
{
free(kmeans_data.means[i].key);
free(kmeans_data.means[i].val);
}
free (kmeans_data.means);
free (means);
free(kmeans_data.clusters);
get_time (&end);
#ifdef TIMING
fprintf (stderr, "finalize: %u\n", time_diff (&end, &begin));
#endif
return 0;
}
/**
* This application groups 'num_points' row-vectors (which are randomly
* generated) into 'num_means' clusters through an iterative algorithm - the
* k-means algorith
*/
int main(int argc, char **argv)
{
int **points;
int **means;
int *clusters;
int i;
struct timeval begin, end;
parse_args(argc, argv);
points = (int **)malloc(sizeof(int *) * num_points);
for (i=0; i<num_points; i++)
{
points[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating points\n");
generate_points(points, num_points);
means = (int **)malloc(sizeof(int *) * num_means);
for (i=0; i<num_means; i++)
{
means[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating means\n");
generate_points(means, num_means);
clusters = (int *)malloc(sizeof(int) * num_points);
memset(clusters, -1, sizeof(int) * num_points);
modified = true;
dprintf("\n\nStarting iterative algorithm\n");
get_time (&begin);
while (modified)
{
modified = false;
dprintf(".");
find_clusters(points, means, clusters);
calc_means(points, means, clusters);
}
get_time (&end);
#ifdef TIMING
fprintf (stderr, "library: %u\n", time_diff (&end, &begin));
#endif
dprintf("\n\nFinal Means:\n");
dump_matrix(means, num_means, dim);
dprintf("Cleaning up\n");
for (i=0; i<num_means; i++) {
free(means[i]);
}
free(means);
for (i=0; i<num_points; i++) {
free(points[i]);
}
free(points);
return 0;
}
int main(int argc, char **argv)
{
int num_points; // number of vectors
int num_means; // number of clusters
int dim; // Dimension of each vector
int grid_size; // size of each dimension of vector space
int num_procs, curr_point;
int i;
pthread_t pid[256];
pthread_attr_t attr;
int num_per_thread, excess;
num_points = DEF_NUM_POINTS;
num_means = DEF_NUM_MEANS;
dim = DEF_DIM;
grid_size = DEF_GRID_SIZE;
{
int c;
extern char *optarg;
extern int optind;
while ((c = getopt(argc, argv, "d:c:p:s:")) != EOF)
{
switch (c) {
case 'd':
dim = atoi(optarg);
break;
case 'c':
num_means = atoi(optarg);
break;
case 'p':
num_points = atoi(optarg);
break;
case 's':
grid_size = atoi(optarg);
break;
case '?':
printf("Usage: %s -d <vector dimension> -c <num clusters> -p <num points> -s <grid size>\n", argv[0]);
exit(1);
}
}
if (dim <= 0 || num_means <= 0 || num_points <= 0 || grid_size <= 0) {
printf("Illegal argument value. All values must be numeric and greater than 0\n");
exit(1);
}
}
printf("Dimension = %d\n", dim);
printf("Number of clusters = %d\n", num_means);
printf("Number of points = %d\n", num_points);
printf("Size of each dimension = %d\n", grid_size);
int ** points = (int **)malloc(sizeof(int *) * num_points);
for (i=0; i<num_points; i++)
{
points[i] = (int *)malloc(sizeof(int) * dim);
}
dprintf("Generating points\n");
generate_points(points, num_points, dim, grid_size);
int **means;
means = (int **)malloc(sizeof(int *) * num_means);
for (i=0; i<num_means; i++)
{
// means[i] = (int *)malloc(sizeof(int) * dim);
means[i] = (int *)malloc(128);
}
dprintf("Generating means\n");
generate_points(means, num_means, dim, grid_size);
int * clusters = (int *)malloc(sizeof(int) * num_points);
memset(clusters, -1, sizeof(int) * num_points);
pthread_attr_init(&attr);
pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM);
CHECK_ERROR((num_procs = sysconf(_SC_NPROCESSORS_ONLN)) <= 0); // FIX ME
//CHECK_ERROR((num_procs = 4 * sysconf(_SC_NPROCESSORS_ONLN)) <= 0); // FIX ME
// CHECK_ERROR( (pid = (pthread_t *)malloc(sizeof(pthread_t) * num_procs)) == NULL);
int modified = true;
printf("Starting iterative algorithm!!!!!!\n");
/* Create the threads to process the distances between the various
points and repeat until modified is no longer valid */
int num_threads;
thread_arg arg[256];
while (modified)
{
num_per_thread = num_points / num_procs;
excess = num_points % num_procs;
modified = false;
dprintf(".");
curr_point = 0;
num_threads = 0;
while (curr_point < num_points) {
// CHECK_ERROR((arg = (thread_arg *)malloc(sizeof(thread_arg))) == NULL);
arg[num_threads].start_idx = curr_point;
arg[num_threads].num_pts = num_per_thread;
arg[num_threads].dim = dim;
arg[num_threads].num_means = num_means;
arg[num_threads].num_points = num_points;
arg[num_threads].means = means;
arg[num_threads].points = points;
arg[num_threads].clusters = clusters;
if (excess > 0) {
arg[num_threads].num_pts++;
excess--;
}
curr_point += arg[num_threads].num_pts;
num_threads++;
}
// printf("in this run, num_threads is %d, num_per_thread is %d\n", num_threads, num_per_thread);
for (i = 0; i < num_threads; i++) {
CHECK_ERROR((pthread_create(&(pid[i]), &attr, find_clusters,
(void *)(&arg[i]))) != 0);
// EDB - with hierarchical commit we would not have had to
// "localize" num_threads.
}
assert (num_threads == num_procs);
for (i = 0; i < num_threads; i++) {
int m;
pthread_join(pid[i], (void *) &m);
modified |= m;
}
num_per_thread = num_means / num_procs;
excess = num_means % num_procs;
curr_point = 0;
num_threads = 0;
assert (dim <= MAX_DIM);
// printf("in this run again, num_threads is %d, num_per_thread is %d\n", num_threads, num_per_thread);
while (curr_point < num_means) {
// CHECK_ERROR((arg = (thread_arg *)malloc(sizeof(thread_arg))) == NULL);
arg[num_threads].start_idx = curr_point;
// arg[num_threads].sum = (int *)malloc(dim * sizeof(int));
arg[num_threads].num_pts = num_per_thread;
if (excess > 0) {
arg[num_threads].num_pts++;
excess--;
}
curr_point += arg[num_threads].num_pts;
num_threads++;
}
for (i = 0; i < num_threads; i++) {
CHECK_ERROR((pthread_create(&(pid[i]), &attr, calc_means,
(void *)(&arg[i]))) != 0);
}
// printf ("num threads = %d\n", num_threads);
// printf ("num procs = %d\n", num_procs);
assert (num_threads == num_procs);
for (i = 0; i < num_threads; i++) {
pthread_join(pid[i], NULL);
// free (arg[i].sum);
}
}
dprintf("\n\nFinal means:\n");
//dump_points(means, num_means, dim);
for (i = 0; i < num_points; i++)
free(points[i]);
free(points);
for (i = 0; i < num_means; i++)
{
free(means[i]);
}
free(means);
free(clusters);
return 0;
}
// -----------------------------------------------------------------------------
//
// -----------------------------------------------------------------------------
void SampleSurfaceMesh::execute()
{
setErrorCondition(0);
dataCheck();
if(getErrorCondition() < 0) { return; }
DataContainer::Pointer sm = getDataContainerArray()->getDataContainer(m_SurfaceMeshFaceLabelsArrayPath.getDataContainerName());
SIMPL_RANDOMNG_NEW()
#ifdef SIMPLib_USE_PARALLEL_ALGORITHMS
tbb::task_scheduler_init init;
bool doParallel = true;
#endif
TriangleGeom::Pointer triangleGeom = sm->getGeometryAs<TriangleGeom>();
// pull down faces
int64_t numFaces = m_SurfaceMeshFaceLabelsPtr.lock()->getNumberOfTuples();
// create array to hold bounding vertices for each face
FloatArrayType::Pointer llPtr = FloatArrayType::CreateArray(3, "_INTERNAL_USE_ONLY_Lower_Left");
FloatArrayType::Pointer urPtr = FloatArrayType::CreateArray(3, "_INTERNAL_USE_ONLY_Upper_Right");
float* ll = llPtr->getPointer(0);
float* ur = urPtr->getPointer(0);
VertexGeom::Pointer faceBBs = VertexGeom::CreateGeometry(2 * numFaces, "_INTERNAL_USE_ONLY_faceBBs");
// walk through faces to see how many features there are
int32_t g1 = 0, g2 = 0;
int32_t maxFeatureId = 0;
for (int64_t i = 0; i < numFaces; i++)
{
g1 = m_SurfaceMeshFaceLabels[2 * i];
g2 = m_SurfaceMeshFaceLabels[2 * i + 1];
if (g1 > maxFeatureId) { maxFeatureId = g1; }
if (g2 > maxFeatureId) { maxFeatureId = g2; }
}
// add one to account for feature 0
int32_t numFeatures = maxFeatureId + 1;
// create a dynamic list array to hold face lists
Int32Int32DynamicListArray::Pointer faceLists = Int32Int32DynamicListArray::New();
std::vector<int32_t> linkCount(numFeatures, 0);
// fill out lists with number of references to cells
typedef boost::shared_array<int32_t> SharedInt32Array_t;
SharedInt32Array_t linkLocPtr(new int32_t[numFaces]);
int32_t* linkLoc = linkLocPtr.get();
::memset(linkLoc, 0, numFaces * sizeof(int32_t));
// traverse data to determine number of faces belonging to each feature
for (int64_t i = 0; i < numFaces; i++)
{
g1 = m_SurfaceMeshFaceLabels[2 * i];
g2 = m_SurfaceMeshFaceLabels[2 * i + 1];
if (g1 > 0) { linkCount[g1]++; }
if (g2 > 0) { linkCount[g2]++; }
}
// now allocate storage for the faces
faceLists->allocateLists(linkCount);
// traverse data again to get the faces belonging to each feature
for (int64_t i = 0; i < numFaces; i++)
{
g1 = m_SurfaceMeshFaceLabels[2 * i];
g2 = m_SurfaceMeshFaceLabels[2 * i + 1];
if (g1 > 0) { faceLists->insertCellReference(g1, (linkLoc[g1])++, i); }
if (g2 > 0) { faceLists->insertCellReference(g2, (linkLoc[g2])++, i); }
// find bounding box for each face
GeometryMath::FindBoundingBoxOfFace(triangleGeom, i, ll, ur);
faceBBs->setCoords(2 * i, ll);
faceBBs->setCoords(2 * i + 1, ur);
}
// generate the list of sampling points from subclass
VertexGeom::Pointer points = generate_points();
if(getErrorCondition() < 0 || NULL == points.get()) { return; }
int64_t numPoints = points->getNumberOfVertices();
// create array to hold which polyhedron (feature) each point falls in
Int32ArrayType::Pointer iArray = Int32ArrayType::NullPointer();
iArray = Int32ArrayType::CreateArray(numPoints, "_INTERNAL_USE_ONLY_polyhedronIds");
iArray->initializeWithZeros();
int32_t* polyIds = iArray->getPointer(0);
#ifdef SIMPLib_USE_PARALLEL_ALGORITHMS
if (doParallel == true)
{
tbb::parallel_for(tbb::blocked_range<size_t>(0, numFeatures),
SampleSurfaceMeshImpl(triangleGeom, faceLists, faceBBs, points, polyIds), tbb::auto_partitioner());
}
else
#endif
{
SampleSurfaceMeshImpl serial(triangleGeom, faceLists, faceBBs, points, polyIds);
serial.checkPoints(0, numFeatures);
}
assign_points(iArray);
notifyStatusMessage(getHumanLabel(), "Complete");
}
int main(int argc, char **argv)
{
int i, j, iter;
int k;
int min_index;
double min_dist;
double *delta;
double *delt_perc;
struct Point *centroids;
struct Point *data;
FILE *file;
srand(time(NULL));
printf("K-Means Test\n");
printf("============\n");
if (argc > 1)
k = atoi(argv[1]); //assume first argument is desired k value
else
k = 3;
centroids = malloc(k * sizeof(struct Point));
data = malloc(DATA_POINTS * sizeof(struct Point));
delta = malloc(k * sizeof(double));
delt_perc = malloc(k * sizeof(double));
generate_points(k, centroids, 0, 0, 0, 0);
generate_points(DATA_POINTS, data, 0, 0, 0, 0);
//label centroids
printf("Initial Centroids\n");
printf("=================\n");
for(i = 0; i < k; i++)
{
printf("{%f, %f}\n", centroids[i].x, centroids[i].y);
centroids[i].label = i;
}
iter = 0;
do
{
char *gnuplot_commands[RULES];
gnuplot_commands[1] = "unset key";
char *plot = malloc(sizeof(char) *(k + 2) * strlen(" 'data_clust.dat' ,") + 20);
strcat(plot, "plot ");
//label all sample data from initial centroids
for(i = 0; i < DATA_POINTS; i++)
{
min_dist = distance(&data[i], ¢roids[0]);
min_index = 0;
int j;
for(j = 1; j < k; j++)
{
double dist = distance(&data[i], ¢roids[j]);
if (dist < min_dist)
{
min_dist = dist;
min_index = j;
}
data[i].label = min_index;
}
}
//write out data points for this iteration
for(i = 0; i < k; i++)
{
char fname[strlen("data_clust.dat") + k];
sprintf(fname, "data_clust%d.dat", i);
file = fopen(fname, "w");
strcat(plot, "'");
strcat(plot, fname);
strcat(plot, "', ");
if(file == NULL)
{
perror("Failed to open data file");
return -1;
}
for(j = 0; j < DATA_POINTS; j++)
{
if(data[j].label == centroids[i].label)
fprintf(file, "%f %f\n", data[j].x, data[j].y);
}
fclose(file);
}
char cenName[strlen("cent.dat") + k];
sprintf(cenName, "cent_%d.dat", i);
file = fopen(cenName, "w");
strcat(plot, "'");
strcat(plot, cenName);
strcat(plot, "' ");
if(file == NULL)
{
perror("Failed to open init_cent.dat");
return -1;
}
//write out centroids
for(i = 0; i < k; i++)
{
fprintf(file,"%f %f\n", centroids[i].x, centroids[i].y);
}
fclose(file);
//plot clusters and centroids
gnuplot_commands[0] = malloc(sizeof(char) * strlen("set title 'K-Means Clustering Iter' ") + iter + 1);
sprintf(gnuplot_commands[0], "set title 'K-Means Clustering Iter %d'", iter);
gnuplot_commands[2] = plot;
//print commands
printf("\n");
for(i = 0; i < RULES; i++)
{
printf("%s\n", gnuplot_commands[i]);
}
FILE *gnuplotPipe = popen("gnuplot -persistent", "w");
for(i = 0; i < RULES; i++)
fprintf(gnuplotPipe, "%s \n", gnuplot_commands[i]);
fclose(gnuplotPipe);
//calculate new centroids based on initial clustering
printf("New Centroids\n");
printf("=============\n");
for(i = 0; i < k; i++)
{
struct Point * mean = get_centroid_mean(¢roids[i],
data, DATA_POINTS);
double diff = distance(¢roids[i], mean);
double change = diff / ((diff + delta[i])/2);
delta[i] = diff;
delt_perc[i] = change;
centroids[i].x = mean->x;
centroids[i].y = mean->y;
free(mean);
printf("{%f, %f}\n", centroids[i].x, centroids[i].y);
}
printf("\n");
// free(plot);
iter++;
}
while(average(delt_perc, k) > THRESHOLD);
free(centroids);
free(data);
free(delta);
free(delt_perc);
return 0;
}
