Whitewater is a proof of concept visual dataflow language for creating GPU accelerated sklearn pipelines. The GUI has been forked from AlvarBer/Persimmon. Currently focused on linear regression, Whitewater substitutes Pandas' read_csv calls with those from RAPIDS cuDF and Linear Regression from cuML.
Input data size is assumed to be MxN, where the first N-1 columns make up your independent variable, X - and the last column (N) is assumed to be your dependent variable, Y.
Visual Demo is located on YouTube
It represents functions as blocks, inputs and outputs are presented as pins, and type safety is enforced when the connection is being made.
A smart bubble helps suggesting suitable context-sensitive blocks when making a connection, showing only the blocks which are type safe. There is also a search box that can be used for finding a particular block.
Whitewater leverages the RAPIDS cuDF, cuML, and cuGraph nightly builds. I recommend using Anaconda as the base package manager.
Install Whitewater dependencies
conda env create --name whitewater --file requirements-cuda10.0.yml
Activate Whitewater environment
conda activate whitewater
Install Whitewater (first time use)
python setup.py build
python setup.py install
To execute use
python -m whitewater
