Auction site Trade Me trials machine learning in the cloud

Microsoft's Azure ML fits the bill to predict and improve online auction outcomes.

Trade Me is dipping its toes in predictive analytics, deploying Microsoft's Azure ML to predict auction results.

The New Zealand auction site has also been able to experiment with the cloud machine learning platform for no license cost so far, using the free version.

"Basically the free version is smaller and slower," Trade Me's senior business intelligence architect, Philip Seamark, told an audience at Microsoft's Ignite conference on Wednesday.

Azure ML delivers production-ready machine learning, Seamark said. No programming is required in the drag and drop interface.

"While as a coder that is hard to accept, some of the challenges machine learning solves are not easy to do while writing code," Seamark said.

The code needed to predict auction results is one example. Seamark said he couldn't imagine the code that would be needed to solve that.

Azure ML requires analytics professionals not necessarily data scientists, who are in short supply, he said. It does, however, provide "first-class" support for the R language and for collaboration.

Trade Me launched its Azure ML project through an internal competition to predict auction outcomes and through that to better understand the value of its premium products. Such a predictive understanding could help Trade Me guide its customers to better online auction results.

Azure ML guides users to the right algorithm and allows them to visually compose the query environment, Seamark said. It can also deliver a production-ready web service at the push of a button.

Being browser based (HTML5), Azure ML is also far more accessible than desktop analytics.

To refine its early competition models, Trade Me's team used multiple models in further experiments and Azure ML's "sweep parameters" module to optimise its parameter settings.

Azure ML's "feature hashing" functionality helped to further increase accuracy by turning text in the auction listings into data that could be used, Trade Me data analyst Simon Carryer said.