LOLA-Voronoi

[jschueller]

implements crombecq2011

algo = otexp.LOLAVoronoi(x0, y0, distribution)
for i in range(5):
    x = algo.generate(15)
    y = f1(x)
    algo.update(x, y)

[sofianehaddad]

Maybe adding a specific test with nanoflann enabled ?

[jschueller]

I can't disable nanoflann at runtime: either nanoflann is enabled at buildtime if available, either its the regular implementation. anyway they give the same results. but both are tested since nanoflann is not available on the mindeps job for example

[sofianehaddad]

ok

[regislebrun]

I have only checked the nanoflann part in details. IMO it should be part of the next release and should be merged, while the remaining of the code is for a longer term.

[regislebrun]

This should be in the experimental module. I fixed it in the commit related to the MarginalUniformOrderStatistics, my bad

[jschueller]

yes, it would be easier to squash the CombinationsDistribution commits

[regislebrun]

These formulas are a copy/paste from KPermutationsDistribution, Anne has to change it into my branch soon

[mbaudin47]

Suggested change

	// Generate new samples
	// Generate a new sample

[m-balesdent]

I see that the SequentialSamplingAlgorithm.cxx is located in Uncertainty\Algorithm\Optimization.

IMO, this algorithm is quite generic and could be the base for other types of goal-oriented algorithms in the future.

For example, AK methods follow the same architecture.

Thus, maybe this base class could be moved outside the Optimization folder to anticipate future developments.