This shutdown had an incredibly healthy discussion internally. The reality is that this service had been unmaintained for a long while, but we'd previously chosen not to start this deprecation process until we had a GA service we could actually have someone migrate to (Cloud ML Engine).<p>Additionally, it turns out that very few people were using it. That's not an excuse, but the reality of ongoing investment. I fought hard for this to be our expected 1 year term, and we had hoped to have a somewhat cookie cutter guide for "Here's how you reproduce this with TensorFlow". Quite frankly, the handful of users of the prediction API likely aren't the kind to happily port to TensorFlow (and this service has existed since the sort of App Engine only days, so they're mostly hobbyists, but I still care).<p>It's never great to "have to" turn down a service, but ultimately when forced between letting the code rot and become a potential security nightmare versus give the small set of users some time to retool, the decision was made to go with the latter. No new features is an easy way to keep something running forever, but keeping the damn thing secure requires a team to stay on top of it.<p>Disclosure: I work on Google Cloud.
Before folks start comparing this to Reader or point to the general "Google shuts things down" narrative, Prediction API has been superset by the array of ML APIs and Google Cloud ML, found at [0].<p>[0] <a href="https://cloud.google.com/products/machine-learning/" rel="nofollow">https://cloud.google.com/products/machine-learning/</a><p>(work at G)
One thing that I find so amazing about AWS (Amazon Web Services) is that I'm not aware of them ever EOLing one of their apis (I could be wrong). We still have a bunch of code that still uses SimpleDB and even though they haven't promoted SDB for a while, they haven't EOLed it.
> Q: What will happen with my existing models?<p>> A: You must recreate your existing Prediction API models using Cloud Machine Learning Engine. To learn more, please read our documentation about creating models on Cloud Machine Learning Engine.<p>Slightly edited and corrected answer should be<p>> A: You must recreate your existing Prediction API models using Cloud Machine Learning Engine. But, you know what? You must recreate it using Amazon Machine Learning, because we might again shut down this service in favor of our next platform. So If you care about your product, move directly to Amazon Machine Learning so next time you will not be bothered by us. And thanks for using it
I'm tired of Google shutting down or screwing up services. I come from marketing, and while I still use Adwords, I'm more and more moving to other platforms for research and spending my money. It's not only that, the moment you have a problem, it's up to yourself, while other companies have a customer service that actually replies, with more or less success.<p>I'm also learning programming, basically because I want to get into Data Science and I'm doing everything I can to avoid using Google Cloud, even though I found Bigtable to be easy to use and the kind of stuff that I wanted for a project, but I forced myself into learning how to get a postgre & couch dbs up. Also using vps's from a local provider (clouding.io).<p>It's like... I can see myself in the future spending time on modifying stuff because Google make X decision, instead of doing stuff I enjoy.
The discussion seems to have pretty well settled on Google's policies around service deprecation, but in case anyone's interested in chatting about the replacement API, I'm excited about the long-term prospects of their ML Engine product.<p>The open-source distributed tensorflow stuff is pretty nice, but it still requires a huge amount of hand coding and tuning the machinery, reminding me quite a lot of just rolling the damn thing in MPI yourself. I'm very excited to see where distributed tf will be in a year or two, but it's a chore today.<p>The hope is that using Google's secret sauce to auto-distribute the execution graphs and associated data ingestion makes things "just work". At the moment, the documentation and examples for that are a bit all over the place at the moment, and require writing models to conform to the newish tf.contrib.learn.Experiment API, which is also a bit underdocumented and underexampled. Using it for very large datasets (say >tens of TB) seems to be pretty challenging at this moment (to me at least).<p>At any rate, I've been banging around on it for a few weeks and am really hopeful. I will follow Cloud ML Engine's career with considerable interest.
tl;dr Cloud Prediction is deprecated in favor of Cloud Machine Learning (<a href="https://cloud.google.com/ml-engine/" rel="nofollow">https://cloud.google.com/ml-engine/</a>)
Now this is going to be a moneymaker, this is a sensible business move.<p>People misunderstand Google: they offer things for free because it lets them collect data (search, gmail, etc.) or it drives competition out (1TB free for BigQuery, for example).<p>For everything else, they either shut them down or charge money =)