void Minimizer_minimizeManyTimes (Minimizer me, long numberOfTimes, long maxIterationsPerTime, double tolerance) {
double fopt = my minimum;
int monitorSingle = numberOfTimes == 1;
autoNUMvector<double> popt (NUMvector_copy<double> (my p, 1, my nParameters), 1);
if (! monitorSingle) {
Melder_progress (0.0, L"Minimize many times");
}
/* on first iteration start with current parameters 27/11/97 */
for (long i = 1; i <= numberOfTimes; i++) {
Minimizer_minimize (me, maxIterationsPerTime, tolerance, monitorSingle);
Melder_casual ("Current %ld: minimum = %.17g", i, my minimum);
if (my minimum < fopt) {
NUMvector_copyElements (my p, popt.peek(), 1, my nParameters);
fopt = my minimum;
}
Minimizer_reset (me, 0);
if (! monitorSingle) {
try {
Melder_progress ( (double) i / numberOfTimes, Melder_integer (i), L" from ",
Melder_integer (numberOfTimes)); therror
} catch (MelderError) {
Melder_clearError (); // interrurpt, no error
break;
}
}
}
if (! monitorSingle) {
Melder_progress (1.0, 0);
}
Minimizer_reset (me, popt.peek());
}
static void _FFNet_Pattern_Activation_learn (FFNet me, Pattern pattern,
Activation activation, long maxNumOfEpochs, double tolerance,
Any parameters, int costFunctionType, int reset) {
try {
_FFNet_Pattern_Activation_checkDimensions (me, pattern, activation);
Minimizer_setParameters (my minimizer, parameters);
// Link the things to be learned
my nPatterns = pattern -> ny;
my inputPattern = pattern -> z;
my targetActivation = activation -> z;
FFNet_setCostFunction (me, costFunctionType);
if (reset) {
autoNUMvector<double> wbuf (1, my dimension);
long k = 1;
for (long i = 1; i <= my nWeights; i++) {
if (my wSelected[i]) {
wbuf[k++] = my w[i];
}
}
Minimizer_reset (my minimizer, wbuf.peek());
}
Minimizer_minimize (my minimizer, maxNumOfEpochs, tolerance, 1);
// Unlink
my nPatterns = 0;
my inputPattern = NULL;
my targetActivation = NULL;
} catch (MelderError) {
my nPatterns = 0;
my inputPattern = 0;
my targetActivation = 0;
}
}
