Add Gaudi algorithm to perform MVA calibration of cluster energies

[giovannimarchiori]

Implement a Gaudi algorithm that reads a model via ONNX runtime and calibrates the clusters
Pre-trained models for ALLEGRO's ECAL will be uploaded in the LAr_scripts package (and FCC-config)

In principle the code could also run a calibration for ECAL+HCAL clusters, if it is trained for that purpose and makes sense.

I put this in RecFCCeeCalorimeter since I'm working on the ee machine but I think the code is general enough that could be promoted to RecCalorimeter.

Also, the code calculates in input the cluster energies per layers from the cells. Since this information is needed also by other algorithms, it would make sense as a next step to factorise this part into a separate algorithms that decorates the clusters in memory with some additional variables (energy fractions, shower shapes, ..) that are available to all subsequent algorithms.

Tagging @BrieucF

[giovannimarchiori]

Hi @kjvbrt ,
thanks for your careful review!
I implemented all your suggestions in the latest commit
cheers,
Giovanni

[BrieucF]

This is really great to have! Thanks Giovanni!

For the trained model, I propose to centralize them here /eos/project/f/fccsw-web/www/filesForSimDigiReco/ALLEGRO/ALLEGRO_o1_v02/, following the same versioning/path conventions as in k4geo

[giovannimarchiori]

Hi @BrieucF , wouldn't it be better to put the files in GitHub, so that we use its versioning system, record changes in the commit messages, can track more easily when files were updated, and so on? They are small files, they are not going to make the repository too big

[BrieucF]

If they are small enough why not

[giovannimarchiori]

Those from LGBM, which has a performance very similar to XGB, are 300-400k. Those from XGB are about 2.5M

[BrieucF]

On the other hand the diff of those files will always be a "full diffs" so not sure the gitHub versioning really brings something more than having our models as model_v1, model_v2, etc. It will pollute the gitHub history and if we introduce bigger models at some point we will end up with heterogenous storage places. NB: the files from /eos/project/f/fccsw-web/www/filesForSimDigiReco/ALLEGRO/ALLEGRO_o1_v02/ can be retrieved with wget or curl, e.g.:

curl -O -L http://fccsw.web.cern.ch/fccsw/filesForSimDigiReco/ALLEGRO/ALLEGRO_o1_v02/elecNoise_ecalBarrelFCCee_theta.root

[BrieucF]

But ok, we can decide later, this should not prevent us to merge this PR