Use standard initialization for PSO instead of full real space sampling

[jrapin]

Types of changes

• Docs change / refactoring / dependency upgrade
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Motivation and Context / Related issue

PSO initialization was very different from other algorithm: sampling in [0,1] and transforming to the real space with an arctan, instead of using a normal distribution. With this diff, PSO uses the sampling of the parametrization, which by default is a normal distribution, like other algorithms.
Closes #296

How Has This Been Tested (if it applies)

Checklist

• The documentation is up-to-date with the changes I made.
• I have read the CONTRIBUTING document and completed the CLA (see CONTRIBUTING).
• All tests passed, and additional code has been covered with new tests.

[teytaud]

PSO was performing great on deceptive, the question is whether the new version will be great as well; I can check this for you maybe ?

[jrapin]

PSO was performing great on deceptive, the question is whether the new version will be great as well; I can check this for you maybe ?

Yes it's important to check, I expect PSO will not be as good: if it's good currently, it may have been because the initialization was completely different to other algorithms', so not a fair comparison

[teytaud]

As I'm underwater I propose that we merge and we'll see in the next global runs if PSO is still dominant in categories of problems for which it is dominant - or we really wait a little bit, I have papers to finish :-)

Regarding fair comparisons - if results with the new initialization are worse, then maybe we must keep a version with the old initialization somewhere.

[jrapin]

@teytaud now:

• PSO uses standard initialization
• WidePSO uses the old initialization
• all chainPSO use the standard PSO (with new initialization)

Should I modify all chainPSO to use the "old" initialization with WidePSO?

[jrapin]

Should I modify all chainPSO to use the "old" initialization with WidePSO?

approved as it is by Olivier