“Release code early and often” is a software engineering philosophy that RAPIDS takes to heart. We try to release about every six weeks or so, partly to keep up the pace of feature development, but also so RAPIDS users don’t get stuck on older versions of our software for too long.
With a six-week release cadence, it’s inevitable that some features we plan to deliver in the next version don’t get finished in time. We hate it when that happens! But, progress continues regardless, and for anyone following along on GitHub, you can track the status of new feature development and bug fixes directly. Want to see what we’re working on for the current release cycle? Check out our project boards: cuDF, cuML, cuGraph.
Want to use a nice new feature or need the relief of a bugfix that just merged? As with all open source projects, if you’re willing to put in the effort, you can always build the current development branch from source. While that’s an ok approach for some, it’s not ideal for everybody. If you’re like many of us in the data science and analytics community, you just want to get your work done, not mess with the vagaries of building and installing from source.
Thus, the RAPIDS devops team has been working hard on producing both nightly RAPIDS containers and conda packages! Now if you wanted to, say, try out cuDF 0.8’s recently merged GPU accelerated JSONlines reader, that’s just a docker pull or conda install command away.
Fun Fact: RAPIDS developers contribute from around the globe, so it’s never “nighttime” for everybody at once. So instead of using an arbitrary local time to build and publish artifacts, our “nightly” conda packages are actually updated each time a PR merges.
If you’re a lover of fast Python and data science, it just got easier for you to try the latest RAPIDS bits. All you need to get started is a Pascal or newer nvidia GPU and Docker or conda.
[Related Article: Self-Driving Cars, Generated News Among Top October Research]