int main() {
unsigned int num_iterations = 1000; // number of iterations to run
float v[2] = {0.0405f, 0.5f}; //
gradsearch gs = gradsearch_create(NULL,v,2,gserror,LIQUID_OPTIM_MINIMIZE);
// execute search one iteration at a time
unsigned int i;
float rmse;
for (i=0; i<num_iterations; i++) {
rmse = gserror(NULL,v,2);
gradsearch_step(gs);
if (((i+1)%100)==0)
gradsearch_print(gs);
}
gradsearch_destroy(gs);
// print results
for (i=0; i<41; i++)
printf(" z = %12.8f, g = %12.8f (%12.8f)\n", z[i], lngamma_test[i], sandbox_lngammaf(z[i], v));
printf("rmse = %12.4e;\n", rmse);
printf("v0 = %12.8f\n", v[0]);
printf("v1 = %12.8f\n", v[1]);
printf("done.\n");
return 0;
}
int main() {
unsigned int num_parameters = 8; // search dimensionality
unsigned int num_iterations = 100; // number of iterations to run
float target_utility = 0.01f; // target utility
float v[num_parameters]; // optimum vector
// ... intialize v ...
// create gradsearch object
gradsearch gs = gradsearch_create(NULL,
v,
num_parameters,
&myutility,
LIQUID_OPTIM_MINIMIZE);
// execute batch search
gradsearch_execute(gs, num_iterations, target_utility);
// clean it up
gradsearch_destroy(gs);
}
int main() {
// options
unsigned int num_samples = 400; // number of samples
float sig = 0.1f; // noise variance
unsigned int num_iterations = 1000; // number of iterations to run
float v[3] = {1, 1, 1};
unsigned int i;
// range
float xmin = 0.0f;
float xmax = 6.0f;
float dx = (xmax - xmin) / (num_samples-1);
// generate data set
float x[num_samples];
float y[num_samples];
for (i=0; i<num_samples; i++) {
x[i] = xmin + i*dx;
y[i] = sincf(x[i]) + randnf()*sig;
}
struct gsdataset q = {x, y, num_samples};
// create gradsearch object
gradsearchprops_s gsprops;
gradsearchprops_init_default(&gsprops);
gsprops.delta = 1e-6f; // gradient approximation step size
gsprops.gamma = 0.002f; // vector step size
gsprops.alpha = 0.1f; // momentum parameter
gsprops.mu = 0.999f; // decremental gamma paramter (best if not exactly 1.0)
gradsearch gs = gradsearch_create((void*)&q, v, 3, gserror, LIQUID_OPTIM_MINIMIZE, &gsprops);
float rmse;
// execute search
//rmse = gradsearch_run(gs, num_iterations, -1e-6f);
// open output file
FILE*fid = fopen(OUTPUT_FILENAME,"w");
fprintf(fid,"%% %s : auto-generated file\n", OUTPUT_FILENAME);
fprintf(fid,"clear all;\n");
fprintf(fid,"close all;\n");
// execute search one iteration at a time
fprintf(fid,"u = zeros(1,%u);\n", num_iterations);
for (i=0; i<num_iterations; i++) {
rmse = gserror((void*)&q,v,3);
fprintf(fid,"u(%3u) = %12.4e;\n", i+1, rmse);
gradsearch_step(gs);
if (((i+1)%100)==0)
gradsearch_print(gs);
}
// print results
printf("\n");
gradsearch_print(gs);
printf(" c0 = %12.8f, opt = 1\n", v[0]);
printf(" c1 = %12.8f, opt = 0\n", v[1]);
printf(" c2 = %12.8f, opt = 1\n", v[2]);
printf(" rmse = %12.4e\n", rmse);
fprintf(fid,"figure;\n");
fprintf(fid,"semilogy(u);\n");
fprintf(fid,"xlabel('iteration');\n");
fprintf(fid,"ylabel('error');\n");
fprintf(fid,"title('gradient search results');\n");
fprintf(fid,"grid on;\n");
// save sampled data set
for (i=0; i<num_samples; i++) {
fprintf(fid," x(%4u) = %12.8f;\n", i+1, x[i]);
fprintf(fid," y(%4u) = %12.8f;\n", i+1, y[i]);
fprintf(fid," y_hat(%4u) = %12.8f;\n", i+1, gsfunc(x[i],v));
}
fprintf(fid,"figure;\n");
fprintf(fid,"plot(x,y,'x', x,y_hat,'-');\n");
fprintf(fid,"xlabel('x');\n");
fprintf(fid,"ylabel('f(x)');\n");
fprintf(fid,"grid on;\n");
fprintf(fid,"legend('data','fit',1);\n");
fclose(fid);
printf("results written to %s.\n", OUTPUT_FILENAME);
gradsearch_destroy(gs);
return 0;
}
