-
Notifications
You must be signed in to change notification settings - Fork 2
Home
Andrew Carnes edited this page Apr 14, 2016
·
10 revisions
This is a BDT package written in C++ that currently depends upon ROOT, a particle physics data analysis package. It was designed to improve the momentum assignment in the CSCTF, a subsystem of a particle physics detector at the Large Hadron Collider.
- Disentangle from ROOT: Need something to read/write XML besides ROOT. Make ROOT XML standalone and include it?
- Also need to replace TMath, TRandom, etc to get rid of ROOT dependence
- Parallelize the code
- Change event data structure
- Check to see if the BDT outperforms TMVA on the same dataset
- Compare running times for training and evaluation to TMVA on the same dataset w/ different number of vars
- Is it possible to speed up the training, besides parallelization?
- Where does the large(?) memory usage come from? Not sure how this compares to other algorithms.
- Write up some intro documentation about BDTs in general
- Test the binary classification and huber loss functions for correctness
- Integrate custom loss functions into TMVA