This repo outlines the dataset generation, simple model definition and some evaluation functions for a track to vertex classifier. It acts as a starting point for the ML@L1 hackathon and will look to implement a continual ML model using the avalanche framework and pytorch

Some dataset exist here: https://cernbox.cern.ch/index.php/s/9P2Qw1ssGcld3Pz

OldKF_test.root 16K events

GTT_TrackNtuple_FH_oldTQ.root 300K events

With a standard install of a anaconda environment

Environment will need ~ 2.2 GB free space

conda env create -f avalanche-environment.yml

source env.sh

To install miniconda to setup an anaconda environment

This installation will need ~ 10 GB free space

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

bash Miniconda3-latest-Linux-x86_64.sh

conda env create -f avalanche-environment.yml

dataset/ contains the functiions for generating datasets from root files as well as default location for Train, Test and Val directories for training and evaluating

model/ contains the model class definition, a simpleNN example is given. As well as default location for SavedModels where trained models are saved

eval/ contains the eval_funcs used for evaluating trained models. As well as default location for plots where performance plots are saved

continualML/ contains a simple avalanche example

To generate datasets use in the dataset dir:

python dataset_generator.py path/to/rootfile.root

For large datasets this will take a while

To run a simple training of the simpleNN model use

python train.py

To evaluate a model use

python eval.py

Email [email protected] for questions