Installing zamba

Zamba has been developed and tested on macOS and Ubuntu Linux for both CPU and GPU configurations.

When a user installs zamba that user must specify to install the GPU or CPU version. If the user fails to make this specification, no version of tensorflow will be installed, thus everything will fail.

Prerequisites

  • Python 3.6
  • ffmpeg
  • xgboost

Python 3.6

We recommend Python installation using Anaconda for all platforms, for more information about how to install Anaconda, here are some useful YouTube videos of installation on different platforms:

FFMPEG version 4.0

FFMPEG is an open source library for loading videos of different codecs, and using ffmpeg means that zamba can be flexible in terms of the video formats we support. FFMPEG can be installed on all different platforms, but requires some additional configuration depending on the platform. Here are some videos/instructions walking through installation of FFMPEG on different platforms:

XGBOOST 0.71

XGBoost is a library for gradient boosting trees, which is often used in ensembled machine learning architectures like zamba. XGBoost may require extra steps on your platform. See below:

XGBoost on Windows

  • Download precompiled xgboost for Python 3.6 and either 32 or 64 bit depending on your version of windows
  • Open a command prompt. If you installed Anaconda, you will want to use an Anaconda command prompt: Start > Anaconda3 > Anaconda Prompt
  • cd Downloads - change directories to your download folder where the precompiled binary is
  • pip install xgboost-0.71-cp36-cp36m-win_amd64.whl - your filename may be different based on the version of Windows

XGBoost on Linux and macOS

XGBoost should install with zamba automatically. If you see a problem with xgboost when installing zamba, the easiest fix is to run conda install xgboost==0.71 -c conda-forge in an Anaconda prompt.

Install Hardware Specific Version of Zamba

zamba is much faster on a machine with a graphics processing unit (GPU), but it has been developed and tested for machine with and without GPU(s).

If you are using Anaconda, run these commands from an Anaconda prompt (Start > Anaconda3 > Anaconda Prompt).

GPU

To install for development with Tensorflow for GPU

$ pip install zamba[gpu]

To use a GPU, you must be using an NVIDIA GPU, installed and configured CUDA, and installed and configured CuDNN per their specifications. Once this is done, you can select to install the version of zamba that uses tensorflow compiled for GPU.

CPU

To install for development with Tensorflow for CPU

$ pip install zamba[cpu]

Operating Systems that have been tested

macOS

zamba has been tested on macOS High Sierra.

Linux

zamba has been tested on Ubuntu versions 16 and 17.

Windows

zamba has been tested on Windows 10.