GT Home : : Campus Maps : : GT Directory

Python users of PACE: Meet “Anaconda”

This entry was posted by on Monday, 30 March, 2015 at

We are happy to announce the availability of “Anaconda” python distribution for data analytics and scientific computing. This distribution is a commercial product, which is free for academic use.

https://store.continuum.io/cshop/anaconda/

The default pyhton (2.7) on PACE systems already offers a comprehensive list of scientific software, so here’s a quick pros & cons to help you decide when to use anaconda:

pros of Anaconda
—————
1. Standalone distribution with no particular dependency on a compiler or MPI stack.
2. A long list of supported libraries for python 2.7 and python 3.4 alike (PACE support for python 3.4 is little to none)
3. Fully tested and supported distribution, backed by the creators of numpy and scipy.
4. Very easy to add and upgrade packages using the “conda” package manager. PACE will regularly upgrade all packages to their latest versions.
5. Constitutes an alternative to PACE distribution when some libraries are found to be missing, outdated, or buggy.

cons of Anaconda
—————
1. Lacks MKL optimizations (which requires a separate license). PACE compiled libraries usually come with MKL and may outperform Anaconda, particularly for linear algebra routines and fast Fourier transform.
2. Limited to the libraries supported by the “conda” package manager (although the list is pretty comprehensive). In comparison, PACE distribution can be extended with any given compatible modules and libraries.
3. Cannot be substituted with PACE versions for scientific packages that are compiled with PACE python.

should you decide to give it a try, here’s how to use it:

# Make sure you remove all PACE python module(s)
module rm python        # Or better yet: module purge

# For python 2.7:
module load anaconda2/2.1.0

# For python 3.4:
module load anaconda3/2.1.0

then use “python” as usual.

Please let us know if you find that a library you need is provided by Anaconda but not PACE python, so we can add it. PACE python will continue to be the primary python distribution, hence must continue to offer all of the libraries that you need for your research.

As always, please contact us at pace-support@oit.gatech.edu and let us know if you have any problems. You can also contact mehmet.belgin@oit.gatech.edu directly to leave general feedback about your experience with this distribution.

Happy computing!

Comments are closed.