Introduction to Statistical Learning

(With Applications in Python)

For ICTP Physics Without Frontiers

Description

This repository provides resources for a course based on "Introduction to Statistical Learning with applications in R".

The R labs and exercises are replaced by Python labs and exercises.

The course is designed for the ICTP Physics Without Frontiers programme.

Project Structure

• lectures/slides/ contains the lecture slides in PDF format.
• lectures/latex/ contains the LaTeX/Beamer sources for the slides.
• islpy_python_package/ provides a Python library needed for the labs and exercises (provides data sets and utilities).
• themes/ contains some CSS/JS hacks for consistent look & feel in jupyter notebooks.
• notebooks/ contains jupyter notebooks for labs and exercises
• datasets/ contains the data sets in CSV format (also available via the islpy Python library).

Requirements

The jupyter notebooks for the labs and exercises require a Python >= 3.6 installation with the usual suspects for statistical/machine learning.

• numpy
• pandas, hdf5, pytables
• matplotplib
• seaborn
• statsmodels
• patsy
• scikit-learn
• PyTorch
• TensorFlow
• jupyter

We highly recommend an conda or miniconda installation for this.

The PyTorch installation instructions. Select Python 3.7 and (unless you know what you are doing) None for CUDA.

In addition, the islpwf library provided by this project is required (see isl_python_package/ sub-directory).

The exercises require some jupyter extensions to behave properly, in particular the exercise2 and freeze extensions. These are provided by the jupyter_contrib_nbextensions package:

conda install -c conda-forge jupyter_contrib_nbextensions

Then restart your jupyter server and activate the relevant extensions from the extensions tab of the launch page (you might have to untick the "check compatibility" check box).

We recommend to install the extensions in your user space and then install the CSS/JS patches from the themes/ sub-directory of this project:

jupyter contrib nbextension install --user

If you get warnings about duplicate files when starting jupyter, you might have to do:

jupyter contrib nbextension uninstall --sys-prefix

Then follow the instructions in the README files under themes/ and restart your jupyter server (the instructions assume you are running a modern Linux distribution).