ROOT PPP Topical Meeting: TMVA

Thursday 24 May 2018, 16:00 → 17:45 Europe/Zurich

4/S-030 (CERN)

Danilo Piparo (CERN)

Presents: guilherme, enrico, kim, javi, gerri, stefan, lorenzo, enric, danilo, axel, philippe, jim, pere.

TMVA today: Kim

Observations

- The dataloader can become a mother class, specialised for the "Traditional loader" and the RDF one

- Can we introduce the concept of TMVA dataset, produced by an (custom) action of RDF?

TMVA in the future: Stefan

- Make TMVA a toolkit, factorising it.

- (6) Arrow provides an industry standard for the memory representation of multidimensional array

- (8) The splitting would be very welcome to factor the present implementation of xvalidation

- TMVA could make the difference providing "shells" to represent models created elsewhere and implement de facto the serving.

Conclusions:

This is a proposal to

- Move the focus from training to data (large datasets) loading and inference

- Make TMVA more modular

 

 

16:00 → 16:20 RDataFrame for data ingestion in TMVA

Speaker: Kim Albertsson (Lulea University of Technology (SE))

tmva-rdataframe-for-data-ingestion.pdf

16:20 → 16:40 TMVA in the future and the role of RDataFrame

Speaker: Stefan Wunsch (KIT - Karlsruhe Institute of Technology (DE))

TMVA in the future and the role of TDataFrame.pdf