/**
* Draw a PGM image of the histogram
*/
void REM::drawHistogram(char * filename) {
RImage img;
int i, j;
double maxValue = histogram_max_val(hist);
img.initialize(binDimension, binDimension);
for(i = 0; i < binDimension; i++) {
for (j = 0; j < binDimension; j++) {
img(i, j, RED) = (int) ceil(((double) histogram_get(hist, i, j) / maxValue) * 255);
}
}
img.save(filename);
}
/**
* Calculate the Shannon entropy of this group's relational histogram
*/
double REM::calculateShannonEntropy() {
int i, j;
double totalElems = 0;
double totalBins = binDimension * binDimension;
totalElems = histogram_sum(hist);
if (totalElems != 0) {
for(i = 0; i < binDimension; i++) {
for(j = 0; j < binDimension; j++) {
double e = 0;
double p = (double) (histogram_get(hist, i, j) / totalElems);
if(p > 0) {
e = -p * log2(p);
} else {
e = 0;
}
entropy += e;
}
}
}
else {
/*
* This is a hack for constant intensity regions
*
* Using this, the entropy value computed after normalization
* below is high (close to 1)
*/
entropy = 6.5;
}
/* Abe's term to penalize small groups */
// entropy += (((totalBins) - 1) * ((log2(exp(1))) / (2 * totalElems))) / totalBins;
if(entropy >= 0) {
entropy /= (double) (log2(totalBins));
}
return entropy;
}
/**
* Print a matrix text file of the histogram bin values
*/
void REM::printHistogramMatrix(char * filename) {
FILE *fp;
int i, j;
double totalElems = histogram_sum(hist);
if ((fp = fopen(filename, "w")) == NULL) {
fprintf(stderr, "Unable to open file %s to print histogram matrix\n", filename);
exit(-1);
}
for(i = 0; i < binDimension; i++) {
for(j = 0; j < binDimension; j++) {
fprintf(fp, "%2.5f ", (double) (histogram_get(hist, i, j) / totalElems));
}
fprintf(fp, "\n");
}
fprintf(fp, "\n");
fclose(fp);
}
/**
* Calculate the Renyi entropy of this group's relational histogram
*/
double REM::calculateRenyiEntropy() {
int i, j;
double totalElems = 0;
double totalBins = binDimension * binDimension;
totalElems = histogram_sum(hist);
if (entropy != 0) entropy = 0;
if (totalElems != 0 && histogram_max_val(hist) != histogram_min_val(hist)) {
for(i = 0; i < binDimension; i++) {
for(j = 0; j < binDimension; j++) {
double e = 0;
double p = (double) (histogram_get(hist, i, j) / totalElems);
if(p > 0) {
e = p * p;
} else {
e = 0;
}
entropy += e;
}
}
} else {
entropy = -1;
return entropy;
}
/* Extra term to penalize small groups - Abe's term */
entropy = -1 * (log2(entropy));
entropy += (( (totalBins) - 1) * ((log2(exp(1))) / (2 * totalElems))) / totalBins;
if(entropy >= 0) {
entropy /= (double) (log2(totalBins));
}
return entropy;
}
int main(int argc, char * argv[])
{
//============================================
// Set input parameters
if (argc < 4|| argc > 4){
printf("Wrong number of inputs. You entered %d. You should enter 4\n.", argc);
myusage();
}
int XLENGTH = atoi(argv[1]);
int YLENGTH = atoi(argv[2]);
int NCLUST = atoi(argv[3]);
int HLENGTH;
double px[XLENGTH];
double py[YLENGTH];
double beta = .00001;
double F;
double infocurve[4] = {1,1,1,1};
//int modus = 0;
// modus = 0 : use p(y|c) cluster centers as initialization
// modus = 1 : use p(c|x) assignment ruls as initialization
double annealingrate = 1.1;
bool plotme = true; //some debug printing
bool debug = false;//verbose debug printing
FILE * histFile;
FILE * probFile;
histogram_t *histo;
histogram_t *pxy;
histogram_t *pygc;
histogram_t *pcgx;
histogram_t *pc;
if(plotme)
{
printf("XLENGTH= %d\n", XLENGTH);
printf("YLENGTH= %d\n", YLENGTH);
printf("NCLUST = %d\n", NCLUST);
}
histFile = fopen(histoName, "r");
if(histFile == NULL)
{
printf("Error: can't access %s\n", histoName);
exit(1);
}
float v;
int j,i = 0;
int y;
int x;
char s [YLENGTH*20];
char * tok;
// //allocate memory for large cluster probability arrays
// double** pygc;
// //allocate pointer memory for first dimension
// pygc = (double**)malloc(YLENGTH*sizeof(double));
// if(NULL == pygc){free(pygc); printf("Memory allocation failed while allocating for pygc[].\n"); exit(-1);}
//
// /*allocate memory for second dimension */
// for(i = 0; i < YLENGTH;i++)
// {
// pygc[i] = (double *) malloc( NCLUST * sizeof(double) );
// if(NULL == pygc[i]){free(pygc[i]); printf("Memory allocation failed while allocating for matrix[x][].\n"); exit(-1);}
// }
//
// double** pcgx;
// //allocate pointer memory for first dimension
// pcgx = (double**)malloc(NCLUST*sizeof(double));
// if(NULL == pcgx){free(pcgx); printf("Memory allocation failed while allocating for pcgx[].\n"); exit(-1);}
//
// /*allocate memory for second dimension */
// for(i = 0; i < NCLUST;i++)
// {
// pcgx[i] = (double *) malloc( XLENGTH * sizeof(double) );
// if(NULL == pcgx[i]){free(pcgx[i]); printf("Memory allocation failed while allocating for matrix[x][].\n"); exit(-1);}
// }
//count how many entries are actually in the histogram file -- these are only the non-zero p(y|x)
while(fscanf(histFile, "%f", &v)!= EOF)
{
i++;
}
rewind(histFile);
if(debug){printf("Counted %d entries in histogram file\n", i);}
//make histogram size 1/2 number entries counted because 1/2 are the indices (y values)
HLENGTH = i/2;
histo = histogram_create(HLENGTH);
x = 0;
//make p(y|x) histogram
//YLENGTH*20 is assuming most words are less than 20 chars long
while(fgets(s, (YLENGTH*20), histFile)!= NULL)
{
tok = strtok(s,"\t");
while(tok != NULL)
{
if(debug){printf("Found word: %d\n", atoi(tok));}
y = atoi(tok);
tok = strtok(NULL, "\t");
if(tok != NULL) //hits null in the word
{
if(debug){printf("Found value: %f\n", atof(tok));}
v = atof(tok);
tok = strtok(NULL, "\t");
histogram_set(histo,x,y,v);
}
}
x++;
}
if(debug)
{
printf("Checking histogram\n");
for(x = 0; x < XLENGTH; x++)
{
for(y = 0; y < YLENGTH; y++)
{
if((v = histogram_get(histo, x, y)))
{
printf("Found doc %d - word %d in histo: %f\n",x,y,v);
}
}
}
}
fclose(histFile);
// now read in the other initial probs
probFile = fopen(probName, "r");
if(probFile == NULL)
{
printf("Error: can't access %s\n", histoName);
exit(1);
}
//*********** p(x) is first line *******************
printf("read in p(x)\n");
fgets(s, (XLENGTH*20), probFile);
//printf("s = %s\n",s);
// printf("length of s = %d\n", strlen(s));
tok = strtok(s,"\t");
x = 0;
while(tok != NULL)
{
//printf("tok = %s\n", tok);
if(strcmp(tok,"\n") != 0)
{
px[x] = atof(tok);
//printf("px(%d) = %f\n", x, px[x]);
x++;
}
tok = strtok(NULL,"\t");
}
//second line is blank
fgets(s, (YLENGTH*20), probFile);
//************** p(y) is third line ************************
fgets(s, (YLENGTH*20), probFile);
printf("read in p(y)\n");
// printf("s = %s\n",s);
// printf("length of s = %d\n", strlen(s));
tok = strtok(s,"\t");
x = 0;
while(tok != NULL)
{
if(strcmp(tok,"\n") != 0)
{
py[x] = atof(tok);
// printf("py(%d) = %f\n", x, py[x]);
x++;
}
tok = strtok(NULL,"\t");
}
//fourth line is blank
fgets(s,(YLENGTH*20), probFile);
//******************** fifth line on is p(xy)
printf("Making p(x;y)\n");
pxy = histogram_create(HLENGTH);
x = 0;
while(fgets(s, (YLENGTH*20), probFile)!= NULL)
{
//get line for doc
//printf("Found %s\n", s);
//split y and p
tok = strtok(s,"\t");
while(tok != NULL)
{
//printf("Found word: %d\n", atoi(tok));
y = atoi(tok);
tok = strtok(NULL, "\t");
if(tok != NULL) //hits null in the word
{
//printf("Found value: %f\n", atof(tok));
v = atof(tok);
tok = strtok(NULL, "\t");
histogram_set(pxy,x,y,v);
}
}
x++;
}
fclose(probFile);
if(debug){
printf("Checking pxy\n");
for(x = 0; x < XLENGTH; x++)
{
for(y = 0; y < YLENGTH; y++)
{
if((v = histogram_get(pxy, x, y)))
{
printf("Found doc %d - word %d in histo: %f\n",x,y,v);
}
}
}
}
//****************** initialize cluster probs *****************************
printf("Making p(c|x)\n");
ini_pcgx(NCLUST, XLENGTH, pcgx);
printf("Now going to anneal\n");
//******************* NOW GO TO ANNEAL **************************
F = anneal(XLENGTH, YLENGTH, NCLUST, beta, pxy, histo, pcgx, pygc, pc,
infocurve, annealingrate, plotme);
printf("Returned from anneal\n");
//cleanup
histogram_destroy(pxy);
histogram_destroy(histo);
histogram_destroy(pygc);
histogram_destroy(pcgx);
histogram_destroy(pc);
// for(i = 0; i < YLENGTH; i++)
// free(pygc[i]);
// free(pygc);
// for(i = 0; i < NCLUST; i++)
// free(pcgx[i]);
// free(pcgx);
exit(1);
}
