Hepynet Example (v0.4.2)

Minimal example to show to work with hepynet v0.4.2

Introduction

Goal of the hepynet: perform DNN related high energy physics analysis tasks with simple config files

• for ATLAS Analysis: include supports for various ATLAS analysis jobs

• Config Driven: all tasks defined by a simple config file

• Python Based: codes are written in Python, which is the mainstream language for DNN studies

This repository setup the workspace to make use of hepynet.

Setup

• Method 1 - With Conda (Recommended)

  conda create -n hepynet python=3.8
  conda activate hepynet
  pip install hepynet

• Method 2 - Direct install (Not recommended)

  pip install hepynet

• Method 3 - With Docker

  Install Docker if not installed.

  Replace the following items in the docker starting scripts:

  • data directory: path to the directory, where you save the input data

  • version: hepynet version, for example v0.4.1

  On every startup

  • Linux Command Line

    source docker/start_docker_example_linux.sh

  • Windows PowerShell

    . docker/start_docker_example_win.bat

  • Windows file explorer

    double-click docker/start_docker_example_win.bat

  Note: if the Docker image is not installed yet, this will automatically pull the required image from Docker Hub

GPU support

• You can refer to Tensorflow GPU support to set up environment to use GPU for training

• This is NOT mandatory, CPU alone is enough to run hepynet

Preparations

• Prepare pandas DataFrame as inputs

  • use uproot to convert ROOT files to pandas DaraFrame, see examples under root_to_pd

• Prepare job config files

  all config files are put under share folder, there are 3 types of config files you should prepare/modify

  • cross_platform.pc_meta.yaml

    You can specify the data folder (where you keep the input files) here

  • train configs: configs for a model training job

  • apply configs: configs for a model applying job

  please refer to config_preparing.md for more details

Example run on lxplus

• Copy example ROOT files

  git clone --recursive git@github.com:Hepynet/hepynet_example.git
  cd hepynet_example
  mkdir -p data/ntuples
  cp /afs/cern.ch/work/y/yangz/public/shared_files/hepynet_example/root_files/*  data/ntuples

  The example inputs are two signal samples at two close mass points (250/300 GeV). One used as signal and the other used as background for usage demonstration.

• Convert ROOT files to hepynet inputs

  follow instructions to convert files

  source root_to_pd/setup_env.sh

Run Hepynet

perform training

conda activate hepynet
hepynet share/example/train_dnn.yaml

applying model to evaluate

hepynet share/example/apply_dnn.yaml