Options for Installing TensorFlow

The installation instructions for TensorFlow are here.

If you are not running Mac or Linux, or if things don't work on your installation of Linux, you might consider installing and dual-booting the latest Linux version (e.g. Linux Mint.) This will give you the best performance. However, if you've never installed Linux before, I would not recommend installing Linux on a machine that you need to use for other tasks in the middle of the semester, because things can go wrong with your existing OS, and figuring out how to fix everything can take time.

A safer option is to run Linux on a virtual machine.

Installing Linux Mint on a Virtual Machine

• Download and install VirtualBox
• Download a Linux Distribution (I recommend the Xfce edition of Linux Mint, available here.)
• Create a new virtual machine in VirtualBox. You may need to enable virtualization technology in your BIOS
• Start your virtual machine, booting using the Linux ISO that you donwloaded, and install Linux

• Once Linux is installed on the virtual machine, if you've installed Linux Mint, use the following to install pip:

  wget https://bootstrap.pypa.io/get-pip.py

  sudo python get_pip.py

• You are now ready to install TensorFlow
• sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.7.0-py2-none-linux_x86_64.whl
• Download and extract Pyzo for Linux. To run it, use [pyzofolder]/pyzo
• You need to switch to Python 2 in Pyzo. The instructions are here

TensorFlow on CDF

Unfortunately, we've had some trouble installing TensorFlow on CDF's Linux. You can, however, run a VM on a CDF workstation (but you cannot do that on wolf) and run TensorFlow inside the VM. To run the VM, run cvm csc321 in the terminal on a CDF workstation. The initial password is Veew1yoo. Pyzo is installed in the VM, and the contents of tf_alexnet are available in the home directory when you run the VM.

Make sure that you're using a machine with > 2GB of RAM when running. The machines with 2GB of RAM are in BA2210 and BA2220, so avoid them. The machines in BA2270 have 4GB of RAM. Try the ones in BA2200, BA3175, BA3185, and BA3195.

TensorFlow on AWS

You can also use AWS to run TensorFlow. There are several community AMIs with TensorFlow pre-installed. TensorFlow runs acceptably fast on t2.medium instances. Here is a video tutorial on using TensorFlow on AWS.