Google has announced that over the course of the next year, it is discontinuing its Cloud Prediction API, and encourages developers using the Prediction API to migrate to its Cloud Machine Learning Engine, which enables data analysis, machine-learning training, and predictions from the Google Cloud Platform. After May 1st, 2018, Google will no longer provide technical support, further releases, or any API access to the Cloud Prediction API.
The Cloud Prediction API provided a RESTful API and machine-learning tools to help developers build machine-learning models to create application features such as recommendation systems, spam detection, and purchase prediction. Its alternative, the Cloud Machine Learning Engine, was opened to the public by Google in March of this year and aims to support many of the same use cases, with enhanced functionality from the TensorFlow machine-learning library.
To migrate to the Cloud Machine Learning Engine, developers must manually recreate any models they had built for the Prediction API in Tensorflow. Developers can find further information on the deprecation and transition steps on the End-of-Life FAQ, which also recommends that developers reach out to Google Cloud Support to discuss approaches to the transition.
As with many product deprecations from Google, this announcement is accompanied by community speculation about the reliability and longevity of Google's service offerings. Solomon Boulos, who works on Google Cloud, weighs in on a recent YCombinator thread on the topic, stating that the API had been "unmaintained for a long while" and that "very few people were using it." Indeed, the announcements forum of the Prediction API’s website has had no activity in over three years, and there has been little development on the product beyond adding a Smart Autofill Add-on for Google Sheets using the Prediction API in late 2014.
Google removed its deprecation policy from the Prediction API in 2014, but the one-year timeline is consistent with the policies of other Google products, and with policies of the same length enforced by comparable companies such as Amazon and Azure. Boulos points out that the Google Cloud team had been waiting to discontinue the Prediction API for a product such as the Cloud Machine Learning Engine to be released, so that developers would have a service to migrate to. Google has published use cases and examples for the Cloud Machine Learning Engine on Github and documentation to help developers get started.