Google Prediction API

This has been around for several years. I tried to get a large company I worked at to use it to complement our internal machine learning algos.<p>We decided not to send data to google because we weren&#x27;t sure about how it would be used or stored. I wonder if that was paranoia from management or if anyone else on HN feels the same way about using a third-party for intellectual property initiatives.

I replaced most of our custom models (build on scikit-learn) with this.<p>Pros:<p>- Super simple API, you can be up and running in a few hours<p>- I don&#x27;t need to maintain machines&#x2F;code to do all the training and querying (huge win for us)<p>- Relatively fast<p>- Very limited ways to influence the models it builds<p>Cons:<p>- Like other Google APIs, it 500&#x27;s from time to time (~0.5%) for no good reason<p>- It&#x27;s a big black box, doesn&#x27;t expose much information about the models it builds other than a confusion matrix.<p>- Very limited ways to influence the models it builds<p>If ML is critical to your business and you know what you are doing, you can get better results with a custom stack.

Can someone give a couple of archetypal examples of what a business would use this for?

Can anyone comment on using this vs <a href="http:&#x2F;&#x2F;prediction.io&#x2F;" rel="nofollow">http:&#x2F;&#x2F;prediction.io&#x2F;</a> ?

Non reliability of API life and data .. have kept me away from this .. no matter how good&#x2F;bad it is now .

Can someone give an example of possible in and output? I have no Idea what I can do with his API besides that it is some kind of data analysis

I want Google to put a live demo of this for weather prediction and stock market prediction to demonstrate how well it works :)

Is it bad that five out of six features are completely generic and would be expected of any product, and the only remaining one tells us nothing at all?<p>Pretty sure you could just replace <i>Prediction</i> on this page with <i>your product</i>.