Analysis group meeting

Tuesday Dec 30, 2025, 4:30 PM → 5:30 PM Asia/Krasnoyarsk

Вадим Костюхин, Ирина Шрaйбер (TSU), Мария Диденко

Link to telemost

For the moment I see only Neda's report

Neda:

1) Sevaral TMVA methods demonstrate overtraining in signal (see e.g. table p.23).  Signal statistics is not enough. I propose to use k-fold training to mitigate this effect

2) Any progress in non-TMVA implemented methods ?

 

Maria:

   Any hope to increase statistics for Neda's study?

1 Мария Диденко

Speaker: Мария Диденко

2 Неда Фироз

Speaker: Неда Фироз (Tomsk State University)

30th Decemeber Updates.pdf

1) Sevaral TMVA methods demonstrate overtraining in signal (see e.g. table p.23).  Signal statistics is not enough. I propose to use k-fold training to mitigate this effect

Answer: Okay I will try to use k-fold training for the affected classifiers and to reassess overtraining using independent test folds

2) Any progress in non-TMVA implemented methods ?

Answer: I was working on python last week, but the unfortunately the Condor experiment ran but with some errors. I will try to fix this issue soon. 

3 Дмитрий Старков

Speaker: Дмитрий Старков