avaneev/biteopt

Derivative-Free Global Optimization Method (C++, Python binding)

biteopt's Issues

wrapParamReal gives parameters outside of range; a typo?

In wrapParamReal, is the line

return( maxv - rnd.get() * ( v - dv ));

supposed to be

return( maxv - rnd.get() * ( v - maxv ));

? This would mirror the symmetric case for when v < minv. In any case, the current code can return a parameter value outside the range. In my test case I have minv: -1000, maxv: 1000, so dv = 2000. The incoming sample is 1039.54, which reaches line 1802, giving 1000 - rnd.get() * (1039.54 - 2000). Since rnd.get() is in [0,1] and the parenthetical is negative, we get a resulting sample greater than 1000, the supposed maximum.

Thanks for biteopt! :)

Memory leak

In biteaux.h and biteopt.h there are "new" statements without corresponding "delete" statements.

For instance in
void updateDims( const int aParamCount, const int M = 6,
const int PopSize0 = 0 )

there is an deleteBuffers() call, but for "new CBiteOptWrap*[ OptCount ];"
there seems no corresponding delete.

This leads to a memory leak in fcmaes (where BitOpt is called here https://github.com/dietmarwo/fast-cma-es/blob/master/_fcmaescpp/biteoptimizer.cpp)

Is there a way I can prevent this leak by adapting the calling code in biteoptimizer.cpp ?

Benchmarking

Hello, I maintain the humpday package which I use to select optimizers. Biteopt was brought to my attention by this issue and I'd like to include it. It seems like the other benchmarking has gone very well.

I'm assuming I can use biteopt? You say there are zero hyper-parameters, which is good, but is there anything obvious I should know?
-Peter