Commit 9a8a4d51 authored by Andrey Filippov's avatar Andrey Filippov

Trying other improvements

parent d76553f1
......@@ -50,6 +50,7 @@ public class SymmVector {
private int [] sym_indices;
private int num_defined;
private boolean [] used_indices;
private double [] cumul_influences;
public int debug_level = -1;
public SymmVector (
......@@ -133,12 +134,13 @@ public class SymmVector {
}
}
boolean use_min_influence = true;
cumul_influences = new double[N];
for (int ivect = 0; ivect <= sym_indices.length; ivect++) { // <= to check for impossible
int best_index = 0;
int num_best0 = -1;
int num_best = -1;
double best_height = -1.0;
double best_metrics = -1.0;
// double best_metrics = -1.0;
if (ivect == 0) {
best_index = 0;
} else {
......@@ -146,7 +148,9 @@ public class SymmVector {
double [] new_mins = use_min_influence ? getNewMins() : null;
// double [] metrics = new_heights.clone();
// best_index = bestVector(metrics);
best_index = bestVector(best_delta, new_heights, new_mins);
// best_index = bestVector(best_delta, new_heights, new_mins);
best_index = bestVector(best_delta, new_heights, best_delta, new_mins, cumul_influences);
if (ivect == 1) {
if (debug_level > -1) System.out.println("Vector # "+ivect+": overwriting best_index= "+best_index+" with 1");
best_index = 1; // overwrite
......@@ -288,6 +292,10 @@ public class SymmVector {
for (int j = 0; j < 2 *N; j++) {
dvectors[indx][j] *= scale;
}
double [] ni = getNormInfluence(dvectors[indx]);
for (int j = 0; j < N; j++) {
cumul_influences[j] += ni[j];
}
}
private double[] remove_projection(double [] new_vect, double [] used_vect) { // |used_vect| === 1.0);
......@@ -396,6 +404,57 @@ public class SymmVector {
}
return num_best;
}
private int bestVector(double delta_primary, double [] primary, double delta_secondary, double [] secondary, double[] cumul_influences) {
if (cumul_influences == null) {
return bestVector(delta_primary, primary, secondary);
}
int ibest = bestVector(primary);
double threshold = primary[ibest] * (1.0 - delta_primary);
boolean [] mask = new boolean [primary.length];
for (int i = 0; i < primary.length; i++) {
mask[i] = primary[i] >= threshold;
}
ibest = bestVector(secondary, mask);
threshold = secondary[ibest] * (1.0 - delta_secondary);
int num_best = 0;
for (int i = 0; i < secondary.length; i++) {
if (mask[i] ) {
if (secondary[i] >= threshold) {
num_best++;
} else {
mask[i] = false;
}
}
}
boolean same_cumul = true;
for (int j = 1; j < N; j++) {
if (cumul_influences[j] != cumul_influences[0]) {
same_cumul = false;
break;
}
}
if (!same_cumul) {
System.out.print("");
}
//Balancing influences does not seem to work - all the remaining have them exactly the same
double [] corr = new double[mask.length];
for (int i = 0; i < mask.length; i++) if (mask[i]){
double [] ni = getNormInfluence(dvectors[i]);
for (int j = 0; j < N; j++) {
corr[i] += cumul_influences[j] * ni[j]; // best has minimal value (most negative)
}
}
int best_index = -1;
for (int i = 0; i < mask.length; i++) if (mask[i]){
if ((best_index < 0) || (corr[i] < corr[best_index])) {
best_index = i;
}
}
return best_index;
}
private int getNumBest(double delta, double [] primary) {
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment