Skip to content

Commit

Permalink
Removed all the print statements used for testing.
Browse files Browse the repository at this point in the history
  • Loading branch information
Jeremy B Karnowski committed Sep 3, 2015
1 parent ac38634 commit 2a1bfae
Showing 1 changed file with 13 additions and 141 deletions.
154 changes: 13 additions & 141 deletions Plugins/Clusterers/GMM/gmm.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,9 +13,8 @@
#include "gmm.h"

// A parallel gmm implementation.
// One thing that I want to improve was to create different ways to initialize the first points.
// The UI interface could choose the initialization method, pass it as a parameter to this function,
// and the initialization code could run differently depending on the parameter value.
// One thing that I want to improve was to create more ways to initialize the first points.

void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned int r, float th, unsigned int meanInit, unsigned int covInit, int *assignment) {

// Parameters for each Multivariate Gaussian Distribution
Expand Down Expand Up @@ -56,7 +55,7 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
// The queue code is the part that makes it parallel.
dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_LOW, 0);

// Initialize all assignments as zero before learning assignments
// Initialize all assignments as zero before learning assignments - ensures does not crash when too many clusters chosen
for(int i = 0; i < n; i++) {
assignment[i] = 0;
}
Expand Down Expand Up @@ -89,8 +88,6 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
// If the covariance is computed from All, then priors are equal
// Otherwise it is calculated from the number of points used in each covariance matrix

printf("Mean = %d, Covariance = %d \n", meanInit, covInit);

switch(meanInit) {

// K-means
Expand All @@ -109,14 +106,11 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
}
means_N[assign] += 1;
}
printf("Counts\n");
for(int i = 0; i < k; i++) {
printf("%d ",means_N[i]);
for(int j = 0; j < d; j++) {
means[i * d + j] /= means_N[i];
}
}
printf("\nK MEANS WORKS!\n");
break;

// Random
Expand All @@ -130,14 +124,6 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
break;
}

printf("MEANS CREATED\n");
for(int i = 0; i < k; i++) {
for(int j = 0; j < d; j++) {
printf("%f ", means[i * d + j]);
}
printf("\n");
}

switch(covInit) {

// Closest
Expand Down Expand Up @@ -166,36 +152,17 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
}
}
});
// printf("Assignments\n");
// for(int i = 0; i < n; i++) {
// printf("%d = %d\n", i, assignment[i]);
// }

// Determine number of points in each covariance matrix
// for(int i = 0; i < n; i++) {
// int j = *(init_assignment + i);
// *(covs_N + j) += 1;
// }

// Determine indices for points in cluster and create covariances
// printf("Number of points: %d\n", n);
for(int i = 0; i < k; i++) {

// int *indices = (int*)malloc(covs_N[i]*sizeof(int));
int count = 0;
// // Determine indices
// Determine indices
for(int j = 0; j < n; j++) {
if(init_assignment[j]==i) {
// indices[count] = j;
count++;
}
}
// printf("Count = %d\n", count);
// printf("Indices:\n");
// for(int jeremy = 0; jeremy < covs_N[i]; jeremy++) {
// printf("%d = %d\n",jeremy,indices[jeremy]);
// }

printf("Computing covar!\n");

// Use the indices to create covariances matrix
covar_indices(data,init_assignment,i,n,d,tmp_cov);
Expand All @@ -206,21 +173,18 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
}
}

printf("Covariance %d\n",i);
for(int l = 0; l < d; l++) {
for(int m = 0; m < d; m++) {
printf("%f ", covariances[i*(d*d) + l*d + m] = tmp_cov[l*d + m]);
}
printf("\n");
}
printf("\n");
// for(int l = 0; l < d; l++) {
// for(int m = 0; m < d; m++) {
// printf("%f ", covariances[i*(d*d) + l*d + m] = tmp_cov[l*d + m]);
// }
// printf("\n");
// }
// printf("\n");

// free(indices);
covs_N[i] = count;
printf("Dist %d number of indices = %f\n", i,covs_N[i]);
}


break;

// All
Expand Down Expand Up @@ -252,18 +216,11 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
// Priors
for(int i = 0; i < k; i++) {
priors[i] = covs_N[i] / n;
printf("%f\n",priors[i]);
}



printf("Starting E-M step\n");

// Until the threshold is reached, perform E-M to update means and covariances
while(fabs(logEstimate - oldLogEstimate) > th) {

//printf("Error = %f \n",fabs(logEstimate - oldLogEstimate));

// Zero out certain arrays
for(int i = 0; i < k; i++) {
*(Nk + i) = 0;
Expand All @@ -273,25 +230,6 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
// function doesn't return anything but it sets covInverses and constantTerms
createpdfs(means, covariances, covInverses, constantTerms, k, d);

// Printing things out to make sure pdf creation worked
printf("Means!\n");
for(int i = 0; i<k; i++) {
for(int j = 0; j<d; j++) {
printf("%f ", means[i*d+j]);
}
printf("\n");
}

printf("Inverse of Covariance\n");
for(int i = 0; i<d; i++) {
for(int j = 0; j<d; j++) {
printf("%f ", covInverses[i*d+j]);
}
printf("\n");
}

printf("E-step!\n");

// E-STEP (compute responsibilities)
//
// This is the parallel version of:
Expand All @@ -302,32 +240,11 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i

dispatch_apply(n, queue, ^(size_t j) {
double min_distance = DBL_MIN;

//double *datum = (double*)calloc(d, sizeof(double));
//double *mn = (double*)calloc(d, sizeof(double));
//double *icv = (double*)calloc(d * d, sizeof(double));
double cnst;
double responsibilitySum = 0;

// Get the point
// for(int i = 0; i < d; i++) {
// datum[i] = data[j * d + i];
// }

// Loop through the possible gaussians and compute responsbilities
for(int l = 0; l < k; l++) {

// get the mean
// for(int f = 0; f < d; f++) {
// mn[f] = means[l * d + f];
// }

// get the inverse of covariance
// for(int q = 0; q < d; q++) {
// for(int r = 0; r < d; r++) {
// icv[q*d + r] = covInverses[l*(d*d) + q*d + r];
// }
// }

// get the constant term
cnst = constantTerms[l];
Expand All @@ -348,49 +265,23 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
}
}


// Divide each responsibility by the sum of all the responsibilities to get posteriors
// Add these to Nk (one value for each gaussian)
for(int l = 0; l < k; l++) {
forLog[j * k + l] = responsibilities[j * k + l];
responsibilities[j * k + l] /= responsibilitySum;
}

//free(distance);
//free(datum);
//free(mn);
//free(icv);
});

printf("Current Assignment!\n");
for(int i = 0; i < n; i++) {
printf("%d ", cur_assignment[i]);
}
printf("\n\n");

// printf("Responsbilities!\n");
// for(int i = 0; i < n; i++) {
// for(int j = 0; j < k; j++) {
// printf("%f ",responsibilities[i * k + j]);
// }
// printf("\n");
// }

// Update Nk - can not be parallelized because it accesses the same data for all j
for(int i = 0; i < n; i++) {
for(int j = 0; j < k; j++) {
Nk[j] += responsibilities[i * k + j];
}
}

printf("M-step!\n");

// M-STEP (recompute gaussians)
dispatch_apply(k, queue, ^(size_t j) {

//double *sample_cov = (double*)calloc(d * d, sizeof(double));
//double *vecMinusMean = (double*)calloc(d, sizeof(double));


// Compute new means - samples times their responsibilities
for(int i = 0; i < n; i++) {
Expand All @@ -402,7 +293,6 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i
means[j * d + l] /= Nk[j];
}


// Compute new covariances
// for each data point
for(int i = 0; i < n; i++) {
Expand All @@ -428,10 +318,7 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i

// Compute new prior
priors[j] = Nk[j]/n;


//free(sample_cov);
//free(vecMinusMean);

});


Expand All @@ -454,27 +341,12 @@ void gmm(float *data, unsigned int n, unsigned int d, unsigned int k, unsigned i

// If this is the best solution so far, copy it to the output.
if ((cur_cost > max_cost) && !isnan(cur_cost)) {
printf("Assignment Changed!\n");
max_cost = cur_cost;
for(int i = 0; i < n; i++) {
assignment[i] = cur_assignment[i];
}
}

printf("Run = %d\n",run);
printf("Current Cost = %f\n",cur_cost);
printf("isnan = %d\n",isnan(cur_cost));

for(int i = 0; i < n; i++) {
printf("%d ", assignment[i]);
}
printf("\n\n");

}



printf("%f\n", cur_cost);

free(tmp_cov);
free(means);
Expand Down

0 comments on commit 2a1bfae

Please sign in to comment.