CERN Latin-American School of High-Energy Physics (CLASHEP)

Introduction

This repository contains jupyter notebooks on statistics and machine learning associated with the lectures at this school.

Dependencies

The notebooks in this package depend on several well-known Python modules, all well-engineered and free!

modules	description
pandas	data table manipulation, often with data loaded from csv files
numpy	array manipulation and numerical analysis
matplotlib	a widely used plotting module for producing high quality plots
imageio	photo-quality image display module
scikit-learn	easy to use machine learning toolkit
pytorch	a powerful, flexible, machine learning toolkit
scipy	scientific computing
sympy	an excellent symbolic mathematics module
iminuit	an elegant wrapper around the venerable CERN minimizer Minuit
emcee	an MCMC module
tqdm	progress bar
joblib	module to save and load Python object
importlib	importing and re-importing modules

Installation

The simplest way to install these Python modules is first to install miniconda (a slim version of Anaconda) on your laptop by following the instructions at:

https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html

I recommend installing miniconda3, which comes pre-packaged with Python 3.

Software release systems such as Anaconda (conda for short) make it possible to have several separate self-consistent named environments on a single machine, say your laptop. For example, you may need to use Python 3.7.5 and an associated set of compatible packages and at other times you may need to use Python 3.9.13 with packages that require that particular version of Python. If you install software without using environments there is the danger that the software on your laptop will eventually become inconsistent. Anaconda (and its lightweight companion miniconda) provide a way, for example, to create a software environment consistent with Python 3.7.5 and another that is consistent with Python 3.9.13. For example, one package may work only with a given version of numpy, while another requires a different version. In principle, having different versions of numpy on your machine, just like having different versions of Python, is not a problem if one uses environments.

Of course, like anything human beings make, miniconda3 is not perfect. There are times when the only solution is to delete an environment and rebuild by reinstalling the desired packages.

Miniconda3

After installing miniconda3, It is a good idea to update conda using the command

conda update conda

Step 1

Assuming conda is properly installed and initialized on your laptop, you can create an environment, here called clashep.

conda create --name clashep

and activate it by doing

conda activate clashep

You need create the environment only once, but you must activate the desired environment whenever you create a new terminal window.

Step 2

Install root, python, numpy, …

	conda install –c conda-forge root

If all goes well, this will install a recent version of the ROOT package from CERN as well as Python and several Python modules including numpy.

Step 3

Install pytorch, matplotlib, scikit-learn, etc.

	conda install –c conda-forge pytorch
	conda install –c conda-forge matplotlib
	conda install –c conda-forge scikit-learn
	conda install –c conda-forge pandas
	conda install –c conda-forge sympy
	conda install –c conda-forge imageio
	conda install –c conda-forge jupyter

Step 4

Install git if it is not yet on your system, then download the CLASHEP package.

	conda install –c conda-forge git
	mkdir tutorials
	cd tutorials
	git clone https://github.com/hbprosper/CLASHEP

In the above the package CLASHEP has been downloaded into a directory called tutorials.

Step 5

Open a new terminal window, navigate to the parent directory that contains CLASHEP, and run the jupyter notebook in that window (in blocking mode).

	jupyter notebook

If all goes well, the jupyter notebook will appear in your default browser. Navigate to the CLASHEP directory and under the Files menu item, click on the notebook test.ipynb and execute it. This notebook tries to import several Python modules. If it does so without error messages, you are ready to try out the other notebooks.

Examples

Statistics

notebook	description
03_rootn	coverage of root(N) upper limits
03_wilks	Wilks' theorem
05_profile_likelihood	calculation of profile likelihood for a signal/background problem

Machine Learning

notebook	description
hzz4l_sklearn	Boosted Decision Trees (BDT) with AdaBoost: classification of Higgs boson events
hzz4l_pytorch	Deep Neural Network (DNN): classification of Higgs boson events
autoencoder1d	Autoencoder: map SDSS galaxy/quasar data to 1D
mnist_cnn	Convolutional Neural Network (CNN): classification of MNIST digits
01_phantom..., 02_phantom..., 03_phantom...	Simulation-Based Inference (SBI): infer parameters of a 2-parameter cosmological model using simulation-based inference. The first notebook generates the simulated Type 1a data; the 2nd notebook performs the simulation-based inference (aka, likelihood-free inference), while the 3rd notebook checks the coverage of the confidence sets. The 2-parameter cosmological model is described in phantom_model.ipynb