Installing Tensorflow using Virtualenv

Introduction

The TensorFlow Python API currently supports Python 2.7 and Python 3.3+ from source.

This tutorial supports a virtualenv of Tensorflow on Mac OS X (10.10.5) and Ubuntu 15.04.

Pip Installs Packages

pip is a package management system used to install and manage software packages written in Python.

Python comes with easy_install (2004). pip is a later alternative (2008), and introduced the idea of requirements files (pip freeze > requirements.txt) which gives users the ability to easily replicate environments (a de-facto POM file for Maven developers). Reference #4 gives a good breakdown on easy_install vs pip.

We can use easy_install to install pip:

sudo easy_install pip

Virtual Environments

virtualenv allows multiple Python projects that have different (and often conflicting) requirements, to coexist on the same computer.

We can use pip to install virtualenv:

$ sudo pip install --upgrade virtualenv

then create the virtual environment:

virtualenv --system-site-packages ~/tensorflow

and finally, activate it:

~/workspaces/public/skflow$ source ~/tensorflow/bin/activate
(tensorflow)~/workspaces/public/skflow$ 

Notice how the session prompt changes to reflect the virtualenv.

To exit the virtualenv at any time, type:

$ (tensorflow)~/workspaces/public/skflow$ deactivate
$ ~/workspaces/public/skflow$ 

Installing Tensorflow into the Virtual Environment

TensorFlow is a C++ library with a Python layer for configuring its internal graph.

Execute either of the following commands within the virtualenv:

Installing on Mac OS X, CPU only:

$ pip install --upgrade https://storage.googleapis.com/tensorflow/mac/tensorflow-0.6.0-py2-none-any.whl

for Ubuntu (GPU enabled):

$ pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.6.0-cp27-none-linux_x86_64.whl

The official documentation will have the latest URLs and versions for installing Tensorflow.

Installing Skflow (Optional)

Skflow is a simplified interface for TensorFlow.

Within the virtualenv, type:

$ pip install -U scikit-learn
$ pip install git+git://github.com/google/skflow.git
$ pip install -U pandas

The github repo for Skflow has some helpful examples. These can be customized to meet common needs and I have found that they provide a quicker on-ramp to productive use of Tensorflow.

Testing the Installation

Run a simple application to ensure the installation was successful.

Sample Python Program:

import tensorflow as tf

a = tf.constant(6)
b = tf.constant(7)

sess = tf.Session()
print(sess.run(a * b))

Desired Output:

(tensorflow)~/workspaces/public/skflow$ python test_tf.py 
I tensorflow/core/common_runtime/local_device.cc:40] Local device intra op parallelism threads: 8
I tensorflow/core/common_runtime/direct_session.cc:58] Direct session inter op parallelism threads: 8
42

References

1. [Sitepoint] Virtual Environments Made Easy
2. [Official Docs] Python Virtualenvs
3. [Blog] Introduction to Pip and Virtual Environments
4. [Official Docs] Pip and easy_install
5. [Official Docs] Tensorflow Docs
6. [Github] Skflow