Methods for Using Anaconda on Personally Identifiable Information (PII)‒Restricted Computers

Author(s):

Dale Shires, Michael Frauenhoffer, and John Vines

Abstract:

The data science, deep learning, and machine learning communities commonly utilize prebuilt and pretested libraries to perform complex operations. Quite often software is written in Python, and libraries are managed with the Anaconda system that allows specific versions of libraries to be integrated with Python easily and quickly. This paradigm, however, gets a bit complicated when the computer system to be used for either model development or inferencing or both is placed behind Internet firewalls and restrictions due to the demands of Personally Identifiable Information processing. All is not lost, however, when using Anaconda and Python in such a trying environment. We discuss two different approaches that solve this problem and give examples of these approaches in action: Singularity containers and Conda-Pack.