Google Prediction API Review

(7 votes, average: 5.00 out of 5)

Loading...

Unfortunately, I disliked Google Prediction API compared to the decent experience using Amazon Machine Learning and wonderful experience using Microsoft Azure Machine Learning.

RobustTechHouse is a leading web & mobile app development company in Singapore focusing on ECommerce, Mobile-Commerce and Financial Technology (FinTech).

Intro

After doing reviews for Amazon Machine Learning and Microsoft Azure Machine Learning, it was suggested to us that we should do a review for the other Machine Learning service on the cloud by the other tech giant, Google, called Google Prediction API and we obliged.

Pricing

These days, free trials are really a must for services on the cloud. Google Prediction API is no exception as you see below. Their free deal (free $300 credit over 60 days) is better than Microsoft Azure Machine Learning’s which is free $200 credit for 30 days. Yay to Google for this.

They also have an aggressive free quota and free usage limits as follows.

But if you start using it for real in production, you have to pay the following prices mentioned at:

https://cloud.google.com/prediction/pricing

Lets try to compare to Microsoft Azure. Both have base monthly fees of $10 which is very affordable. Microsoft Azure API use count is probably comparable to the Google Prediction API Prediction count. In that case both are $0.50/1000 predictions or transactions but Google is cheaper because their first 10,000 predictions are free while Azure is not.

Documentations and Samples

Documentation is lean and can be found at https://cloud.google.com/prediction/docs/developer-guide

Sample templates and hosted models are also very limited compared to Microsoft Azure. Eg I see a few of the following. Nothing compared to the suite available at Microsoft Azure Machine Learning.

Language Identifier:

Analyzes a sentence to determine whether it is English, Spanish, or French.

Tag Categorizer

Tags a given comment as pertaining to android, appengine, chrome, or youtube. Training data comes from a collection of social media comments.

Sentiment Predictor

Analyzes the sentiment of a short English-language text snippet.

Other scenarios and templates are mentioned at https://cloud.google.com/prediction/docs/scenarios

Running Models

Here comes the tough part. Even though we have prior experience with Machine Learning concepts and a few Machine Learning services on the cloud and libraries, we had a hard time figuring out how to actually use Google Prediction API productively (compared to say Amazon and Azure)

We tried using the API Prediction Explorer and with the lack of on-screen guidance, we found it hard to figure out what we are supposed to put into the fields.

Conclusion

Unfortunately, I disliked Google Prediction API compared to the decent experience using Amazon Machine Learning and wonderful experience using Microsoft Azure Machine Learning. We have been doing a bit of work in Machine Learning for a few years. Even if we understood the terminology, it is difficult for us as developers to figure out quickly how to use it efficiently. Business non-tech users probably shouldn’t try this and developers need to be prepared to spend a bit more time than Amazon or Azure to figure out how to use Google’s Prediction API.

There are a few aspects of Google Prediction API we liked including:

• Smart Autofill Spreadsheets as an add-on for doing Machine Learning work directly in Google spreadsheets. That seems useful and interesting. https://cloud.google.com/prediction/docs/smart_autofill_add_on
• Google Prediction API has wide support for different internet languages like Go, Java, Javascript .NET, Node.js, Objective-C, PHP, Python, Ruby via HTTP/JSON

So overall if I have to rank the 3, we like Microsoft Azure Machine Learning the most, followed by Amazon Machine Learning, then we like Google Prediction API the least.