Machine Learning: Google Prediction API
Now that we are all a little more schooled up on the basics of machine learning, it's time to find out who is offering what in the world of off-the-shelf AI.
We'll tackle each of the big four in a four part series in no particular order starting with Google.
Now, Google is predominantly known for Google Search. We use it every day. We use it to find out if our favourite Thai restaurant is open on a Monday night at 7PM (coincidently it is not). We use it to discretely spell check the word necessary whilst no one is looking at work. We use it learn new things, find travel destinations and stalk people we know - a little. We use it for everything, and despite Google Search being one of the most omnipotent tools of the last decade, Google's breadth of technology far exceeds that of Google Search.
This brings us to the Google Cloud Platform. Google Cloud Platform is impressive to say the least. It's build on the exact same infrastructure that Google uses in-house so you know it's going to be fast, scalable and hopefully affordable. The service offerings are vast with everything from computing engines, web storage, SQL storage, no-SQL storage, big data solutions and a swatch of easy-to-use APIs, it's these API, in particular the Prediction API that we are interested in.
Getting started with the Google Cloud Platform couldn't be more simple, already having a Google account means your essentially ready to go. It's probably time we took a quick squiz at pricing and whilst we'll get into more of an in-depth comparison of pricing across the 4 tech giants later, it's probably a good idea to see how much having Google’s servers crunching your number is going to sting. Turns out, not so bad. Google have created somewhat of a care package to get you started, that is $300 worth of credit for use over the course of 60 days - ample to get you started, and what’s more, Google are letting you use the Prediction API for free up to certain limits. All of this makes Google a very enticing option.
We know Google have the services we need. They have storage and they have the Prediction API, but how easy is it to use? This brings us to the last part of the scoping puzzle - documentation. Peering cautiously around the corner toward the documentation and you begin to wonder rather quickly what all the fuss what about. The documentation is comprehensive without being convoluted and it's concise without being boring. You'll instantly feel right at home with the first presentation being a 'Hello Prediction' tutorial.
The API itself is very well done, simple and powerful. It uses a RESTful API which is easy to use and especially handy when you need to constantly train your model on new information. There are also a lot of pre-compiled models so that you can immediately get started analysing customer sentiment, detecting spam, suspicious website activity and more to point churn analysis.
Google have done an outstanding job, setting the bar higher than many of their competitors would probably like. Next up well take a look at how Amazon fares - not exactly known for their UIs and UXs, but given their dedication to R&D could well be the dark horse of the race.