End to End Data Science Without Leaving the GPU

Source materials from Randy Zwitch 'End to End Data Science Without Leaving the GPU' talk at PyData NYC 2018

Build Environment

In order to get all of the packages working in harmony, I built libGDF/pygdf from source, after building pymapd. I've included the conda environment file.

conda create --name pydatanyc2018 --file spec-file.txt

For your best chance at replicating this build environment, using Docker with Ubuntu 16.04 and nvidia-docker2 might be your best bet.

Data

I've included the PJM RTO data created by the df query in cell 3. So to follow along, you could import the example data (exampledata.csv.gz) to pandas, then run the code examples below that query step.