Implementing root TMVA for high level analysis

[juanangp]

Implementing TMVA methods for high level analysis, please check https://root.cern.ch/download/doc/tmva/TMVAUsersGuide.pdf for further details on TMVA.
Summary of changes:

• New class TRestDataSetTMVA to evaluate different TMVA methods on TRestDatasets note that a signal and a background dataSet must be provided together with different TMVA methods. This class evaluate all the methods provided and generates a root file and a folder with the results.
• New class TRestDataSetTMVAClassification to performs the classification of a given TRestDataSet using as input the results of the TMVA evaluation methods generated using TRestDataSetTMVA. An output dataset is generated by
  /// definining a new observable with the TMVA method e.g. BDT_score.
• Added example tmva.rml
• Minor updates on TRestRun and TRestDataSet

[jgalan]

I don't think we should name all the classes using TRestDataSet I think we should assume these classes will use internally TRestDataSet as something natural.

I would only use the DataSet inside the class name when it is clear that we need to differentiate.

For example, inside the sensitivity classes, I plan to have TRestComponent as a base class for TRestDataSetComponent and TRestFormulaComponent.

But it seems to me that TRestDataSetXYZ is over used.

I think discrimination methods usage should be unified by an abstract class TRestDiscrimination. Then TMVA method would be named TRestTMVADiscrimination that uses a TRestDataSet as implemented in TRestDiscrimination.

In my opinion discrimination in our field of research takes such an important role that I would dedicate a directory for all the future discrimination methods. Creating a new directory discrimination.

See issue #13

[juanangp]

I don't think we should name all the classes using TRestDataSet I think we should assume these classes will use internally TRestDataSet as something natural.

Well, I just followed the suggestions done here #392 (comment)
On the other hand, I think we should differenciate between discrimination and data processing, the new TMVA classes are just used to define new observables using TMVA, but the event discrimination is not performed there. I think the TMVA or eithed the LogOdds clasess fits under analysis folder, but we can implement more generic methods for event discrimintation in a different folder.

I don't know how to proceed renaming these classes because is not clear to me the best way to add them to the repository.

[jgalan]

I don't think we should name all the classes using TRestDataSet I think we should assume these classes will use internally TRestDataSet as something natural.

Well, I just followed the suggestions done here #392 (comment) On the other hand, I think we should differenciate between discrimination and data processing, the new TMVA classes are just used to define new observables using TMVA, but the event discrimination is not performed there. I think the TMVA or eithed the LogOdds clasess fits under analysis folder, but we can implement more generic methods for event discrimintation in a different folder.

I don't know how to proceed renaming these classes because is not clear to me the best way to add them to the repository.

Ok, perhaps we should discuss this online