README.md

In joepeskett/kf-tools: A basic package for building dockerfiles and Kubeflow pipeline component in R, with R.

Kubeflow Tools (R)

This is a simple utility package to try and aid with building components for Kubeflow Pipelines in R.

Installation

library(devtools)
install_github('joepeskett/kf-tools')

Note that this will not handle the installation of the required python packages.

Example Usage:

Define an R function:

add <- function(a,b){return(a+b)}

Build a Kubeflow pipeline component yaml

library(kubeflowTools)
component_from_function(add, base_image = 'r:3.6',component_output_file = 'my_first_component.yaml')

Load component when building a pipeline (in Python using the KFP SDK)

import kfp.components as comp
comp.load_component_from_file('my_first_component.yaml)

Depending on how useful this tool is, we may look at building functions for building pipeline packages into this R package also.

Requirements:

R Packages:

• yaml for exporting yaml files
• argparse for formatting the inputs, outputs and arguments to the component.

Python Packages:

• If you want to compile a Kubeflow pipelines package, upload or run a package, you will need to have the kfp package installed, as we will use the command line tools installed alongside the python package.

Docker:

• If you want to build your own docker images to use with the components, then you will need to install docker.

joepeskett/kf-tools documentation built on May 27, 2020, 7:43 a.m.