Advanced Operations: TensorFlow on Jefferson Lab data

Overview

Teaching: 10 min
Exercises: 10 min

Questions

• How do I use Jupyter to go beyond what ROOT can do?

Objectives

• Import a Python library to perform machine learning.

• Train a neural network using gradient descent to find correlations.

Installing your own python packages

Not all packages you need will be installed on the Jefferson Lab Jupyter server. To install your own packages, log in to the interactively farm nodes and make sure you have access to the correct python environment (we will use python 3). To do this, add the python path to your search path:

export PATH=/apps/python/3.4.3/bin:$PATH

or in tcsh

set path = (/apps/python/3.4.3/bin $path)

This will give you access to the pip python package manager. You can get a listing of currently installed packages with

pip list

We first have to setup the proxy so we can access the internet from the interactive farm node. We can do this by setting the following environment variables:

export HTTP_PROXY=http://jprox.jlab.org:8080
export HTTPS_PROXY=https://jprox.jlab.org:8080

or in tcsh

setenv HTTP_PROXY http://jprox.jlab.org:8080
setenv HTTPS_PROXY https://jprox.jlab.org:8080

You can add the commands above to your .login file so they are loaded automatically each time you login.

To install new packages we have to get around the Jefferson Lab proxy, which modifies https traffic. Install (or upgrade if already installed) the package tensorflow with

pip install --trusted-host pypi.org --trusted-host files.pythonhosted.org --user --upgrade tensorflow

When you now restart you python 3 kernel and load tensorflow, you should find the version you just installed.

Key Points

• Use the tensorflow library for machine learning in Python.