A startup called PredictionIO has raised $2.5 million in seed capital to help it try and make a business out of open source machine learning software. Unlike previous open source projects, though, PredictionIO is designed to be easy to get started with and use, even by developers who aren’t data scientists.
PredictionIO claims developers can be writing predictive models for their applications in minutes, primarily it seems around things such as recommendation and personalization. The software is available as a download or as cloud instance on Amazon Web Services. The company itself is part of three startup accelerators — MozillaWebFWD, 500Startups and StartX.
Machine learning is a potentially lucrative software market, and PredictionIO is tackling it by trying to split the difference between open-source and proprietary tools. Open source software is popular — in machine learning that includes projects such as Mahout, scikit-learn and, at some point, Oryx — but often hard to deploy and use. Commercial software is getting much better — with the release of products like GraphLab Create and Microsoft’s new Azure machine learning service — but can be too much like a black box, PredictionIO contends.
However, operating in the middle ground also opens a company up to the risk of fighting competition on two fronts. It could be that developers don’t mind a black-box approach at all, especially as more proprietary tools continue to improve functionality, even into advanced algorithms, and are delivered as cloud services and APIs. It could also be that other open source projects with bigger communities will continue to improve, possibly even spinning out a few startups of their own.
But spending too much time on business concerns now might be putting the cart in front of the horse. The bigger picture here is that thanks to PredictionIO — as well as companies including (but not limited to) GraphLab, Expect Labs, Mortar Data, Microsoft, Wise.io, IBM, Google and AlchemyAPI — developers have more options than ever for using machine learning for building smarter applications, without having to be machine learning experts themselves. Time will tell which approaches to productizing these algorithms will survive, but there’s no putting the machine learning genie back into its bottle.