Updated support for CMA-ES minimizer based on libcmaes

[beniz]

This is an update to PR #40 that includes:

• a rebase with all CMA-ES commits at the tip of master (as of 04/17/2017)
• a set of small fixes as requested by @vgvassilev

As a reminder, this PR fetches, builds and wraps https://github.com/beniz/libcmaes/ with ROOT.

This PR builds ROOT with CMA-ES support and I've been able to run some of the tests, indicating that it is working fine.

Issues
However, some caveats remain, on which help is required, at the moment

• build/include/libcmaes/cmaes.h and build/include/Eigen are not properly passed to the compiler at build time. I cannot find how to do it properly. At the moment I am using symlinks as a temporary hack (see how to build below)

• To access the inner option of the CMA-ES Minimizer, I was using code similar to

ROOT::Math::IOptions &opts = ROOT::Math::MinimizerOptions::Default(fitter);
opts.SetIntValue("lambda",lambda);

Code above now appears to fail with errors such as:

root [0] .L tutorials/fit/cmaesFitBench.C++g
Info in <TUnixSystem::ACLiC>: creating shared library /home/beniz/research/siminole/dev/tmp/root/build/./tutorials/fit/cmaesFitBench_C.so
In file included from input_line_11:9:
././tutorials/fit/cmaesFitBench.C:58:16: error: no type named 'IOptions' in namespace 'ROOT::Math'
   ROOT::Math::IOptions *opts = ROOT::Math::MinimizerOptions::Default(fitter);
   ~~~~~~~~~~~~^
././tutorials/fit/cmaesFitBench.C:58:45: error: no member named 'MinimizerOptions' in namespace 'ROOT::Math'
   ROOT::Math::IOptions *opts = ROOT::Math::MinimizerOptions::Default(fitter);
                                ~~~~~~~~~~~~^
Error in <ACLiC>: Dictionary generation failed!

Help is needed to fix the above.

How to build

cd build
cmake ../ -Dminuit2=on -Dtesting=on -Dlibcmaes=on
make

The build will fail because of the header issue mentioned above, so do:

cd include
ln -s eigen3/Eigen .
ln -s eigen3/unsupported .
cd ..
make

You can then use the newly built ROOT and test that CMA-ES is working:

.L tutorials/fit/cmaesGausFit.C++g
cmaesGausFit()

Please see instructions and links from PR #40 for more tests, performance checks, etc...

Once everything is fine, I'll be able to squash all commits into a single one if needed.

[phsft-bot]

Can one of the admins verify this patch?

[vgvassilev]

@beniz, could you address the coding conventions issues reported by travis?

About the error, it seems you are missing #include "Math/IOptions.h" in cmaesFitBench.C

[beniz]

@vgvassilev FYI I'm pretty busy over the next couple of weeks, if you can test before I fix the coding conventions, please do. It'd be best to fix major issues first. Thanks.

[phsft-bot]

Can one of the admins verify this patch?

[vgvassilev]

I will try to look at it. I am quite busy, too.

[etejedor]

@beniz @lmoneta @vgvassilev, what would you like to do with this PR?

[beniz]

@etejedor Hi, I'd be happy to help get the PR into trunk, I just couldn't find time to struggle with coding conventions.

As of today, I can propose the following:

• I do rebase to current master
• Someone from ROOT helps me through, including to fix coding conventions and anything that can get into the way.

My time is unfortunately scarce, but I'm willing to allocate time to bring this into production.

[oshadura]

@beniz, can you rebase please?

[beniz]

@oshadura Hi, I can, though it will not be immediate, stay tuned, thanks.

[lmoneta]

@oshadura , we don't need to rebase this one now. We need first to test well this PR first.
After that, then we could rebase it.

[beniz]

@lmoneta @oshadura hi, let me know whether I can help you test the PR thoroughly. I'd understand that rebasing could make it easier, depending of the additional tools / testbeds you'd like to use for the testing phase.
I believe libcmaes is sound, so it should really be about the integration into ROOT structures and framework.

[etejedor]

@lmoneta could you provide an answer to @beniz about how to test this PR?

[etejedor]

@lmoneta could you provide an answer to @beniz about how to test this PR?

[lmoneta]

This PR is now replaced by #7044