Build Tensorflow from source, for better performance

build-tensorflow-from-source.md

Building Tensorflow from source on Mac (CPU only) for maximum performance:

TensorFlow is now distributed under an Apache v2 open source license on GitHub.

Step 1. Create new conda environment:

It's a good idea to isolate your dev environment for Tensorflow because:

• It keeps the workspace clean
• Environments can be exported and imported to different machines (In this instance though the point is moot as you'ld be building Tensorflow optimized for your particular system. $ conda create -n <environment name, say tf> python=<Python version of your choice, say 3.6>

Step 2. Activating your environment :

This where you instruct the shell to use your virtual environment. $ conda activate <environment name, say tf>

If this step fails, try running this first: $ source activate <environment name>

Step 3. Install Dependencies:

Conda comes preinstalled with pip and some other basic utils, but the pip version maybe old, so: $ pip install --upgrade pip

Install numpy: $ pip install numpy

Step 4. Install Bazel:

To build TensorFlow from source, the Bazel build system must first be installed as follows. $ brew install bazel

If brew is inot installed: $ cd /usr/local $ mkdir homebrew && curl -L https://github.com/Homebrew/brew/tarball/master | tar xz --strip 1 -C homebrew

Step 5. Install TensorFlow

To obtain the best performance with TensorFlow we recommend building it from source.

First, clone the TensorFlow source code repository:

$ git clone https://github.com/tensorflow/tensorflow
$ cd tensorflow
$ git checkout <version name, say r1.8>

Then run the configure script as follows:

$ ./configure

Output:

Please specify the location of python. [Default is /usr/bin/python]: [enter]
Do you wish to build TensorFlow with Google Cloud Platform support? [y/N] n
No Google Cloud Platform support will be enabled for TensorFlow
Do you wish to build TensorFlow with GPU support? [y/N] n
Configuration finished

Then call bazel to build the TensorFlow pip package:

bazel build -c opt --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --copt=-msse4.2 //tensorflow/tools/pip_package:build_pip_package


bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

This will build the package with optimizations for FMA, AVX and SSE.

And finally install the TensorFlow pip package

For Python 2.7.*:

$ sudo pip install --upgrade /tmp/tensorflow_pkg/tensorflow-*.whl

Python 3.*:

$ sudo pip install --upgrade /tmp/tensorflow_pkg/tensorflow-*.whl

Step 6. Test your installation:

To test the installation, open an interactive Python shell and import the TensorFlow module. Make sure you are not in the tensorflow git repo directory, as that will throw error:

$ cd
$ python

import tensorflow as tf

With the TensorFlow module imported, the next step to test the installation is to create a TensorFlow Session, which will initialize the available computing devices and provide a means of executing computation graphs:

>>> sess = tf.Session()

You should now be able to run a Hello World application:

>>> hello_world = tf.constant("Hello, TensorFlow!")
>>> print sess.run(hello_world)
Hello, TensorFlow!
>>> print sess.run(tf.constant(123)*tf.constant(456))
56088