Overview

Interoperable Coding Practices and Design for Data Science Teams: With Differential Privacy as a first-class object

Initial Goals

• Share Scikit learn models made in python in a python REPL (sharing the python's object sessions) Sharing that object into an R session, and see if one can make predictions based off the "serialized" python module.

• Working proof of concept of rpy2

• Document work being done in blog post and R/Python community feedback [(recieved top post on Rstats)]

• Try out pysyft and pytorch

• Build Atom-based IDE on top of radian (installed terminal and radian)

High Ticket Items:

• Make downloading and setup 1 click

Road Map

• Objects with metadata: Attach an object's history so it can be accessed when transferred

• Document requirements and dependencies in anticipation of creating a R/python virtual environment and/or docker (less priority)??

• Publish pre-print on arxiv

• Utilize apache arrow and parquet to serialize objects for in-memory and on-disk. This would help provide a way to bridge pandas and tidy dataframes.

• Allow easier cloud connections (auth_file locations as environment variables that you have to log into) (single sign on ide)

• Create a dedicated IDE for Federated Learning and Secure Model Communication In Healthcare setting?

• Access objects from either an R or Python process when both are running.