4f84d07 Jun 18, 2018
2 contributors

Users who have contributed to this file

@nioperas06 @martin-gorner
40 lines (33 sloc) 1.79 KB
Python 3 is recommended for this lab. Python 2 works as well if you adapt the installation instructions.
Installation instructions for straightforward pip install below.
If you are a power user under a specific Python environment ((virtualenv, anaconda,
docker), please visit and follow the Python 3 instructions.
# If you do not have it already, install git from
# Install the latest version of python 3 from
# and run this on the command line so that python3 can access https:// URLs:
sudo /Applications/Python\ 3.6/Install\ Certificates.command
# Now you can install tensorflow and matplotlib
pip3 install --upgrade tensorflow
pip3 install --upgrade matplotlib
sudo -H apt-get install git
sudo -H apt-get install python3
sudo -H apt-get install python3-matplotlib
sudo -H apt-get install python3-pip
sudo -H pip3 install --upgrade tensorflow
# you might alo need to upgrade matplotlib, the version pulled by
# apt-get is sometimes stale (but comes with the gfx backend)
sudo -H pip3 install --upgrade matplotlib
Install Anaconda, Python 3 version:
Anaconda comes with matplotlib built in.
In the Anaconda shell type: pip install --upgrade tensorflow
If you get the error "Could not find a version that satisfies the requirement (...)" try the following alternative:
conda config --add channels conda-forge
conda install tensorflow
git clone
cd tensorflow-without-a-phd/tensorflow-mnist-tutorial
=> A window should appear displaying a graphical visualisation and you should also see training data in the terminal.