Sampling interval and update interval should be separate

[GregoryMorse]

It seems that for problems involving trigonometric functions like sine and cosine, the sampling interval should be e.g. (0, 2*PI).

But the update interval should be unbounded. And controlled by using modular arithmetic e.g. fmod(x, 2PI) or fmod(x, 2PI)+2*PI for negative values.

Currently for example, the ADAM local search is clamping the bounds, not using modular arithmetic. See: https://github.com/itsoulos/OPTIMUS/blob/master/src/Optimus/adam.cpp#L29

Should there be a new function Problem::updateparams(vector &x, vector &adjust) which adds the adjustments and custom bounding e.g. with modulo, clamping, etc?

[GregoryMorse]

So I am unclear if bounded input parameters of bounded periodic functions are supported.

My current idea to work around the clamping issue is:

# include <QRandomGenerator>
QRandomGenerator randGen;

void    init(QJsonObject data) {
  if (!init) { randGen.seed(randomSeed); init=true; }
}
void    getmargins(vector<Interval> &x)
{
        for(size_t i=0;i<x.size();i++)
            //x[i]=Interval(0,2*M_PI);            
            //x[i]=Interval(-INFINITY,INFINITY);
            x[i]=Interval(-1e100,1e100);
}

void  getSample(vector<double> &x)
{
    for(size_t i=0;i<x.size();i++)
        x[i]=randGen.generateDouble()*2*M_PI;
}

//f(x)
double  funmin(vector<double> &x)
{
   for (size_t i = 0; i < x.size(); i++) {
       if (x[i] < 0) {
           x[i] = fmod(x[i], 2*M_PI) + 2*M_PI;
       } else if (x[i] > 2 * M_PI) {
           x[i] = fmod(x[i], 2*M_PI);
       }
   }
...
}

[itsoulos]

There is a function in problem called getSample() In there you may use your function for sampling that may be different from the margins!