The official Google Code blog

Author Photo
By Marc Cohen, Developer Relations

This holiday season, the Google Prediction API Team is bringing you four presents and, thanks to the joys of cloud computing, no reindeer are required for delivery. Here’s what you’ve already received:

• Faster on-ramp: We’ve made it easier to get started by enabling you to create an empty model (by sending a trainedmodels.insert request with no storageDataLocation specified) and add training data using the trainedmodels.update method. This change allows you to submit your model contents without needing to stage the data in Google Cloud Storage.
• Improved updates: The algorithms used to implement model updates (adding additional data to existing models) have been modified to work faster than ever.
• More classification algorithms: We’ve increased the number of classification algorithms used to build predictive models, resulting in across-the-board improvements in accuracy.
• Integration with Google Apps Script: Prediction services are now available as part of Google Apps Script, which means you can integrate prediction services with Google Docs, Google Maps, Gmail, and other great Google products.

All of the above enhancements are supported by the current Prediction API version 1.4 so you can enjoy these features using the existing client libraries.

Happy Holidays from the Google Prediction API Team. We’re looking forward to bringing you more exciting features in 2012!


Marc Cohen is a member of Google’s Developer Relations Team in Seattle. When not teaching Python programming and listening to indie rock music, he enjoys using the Google Prediction API to peer into the future.

Posted by Scott Knaster, Editor

Author Photo
By Dan Holevoet, Developer Relations Team

A platform is more than the sum of its component parts. You can read about it or hear about it, but to really learn what makes up a platform you have to try it out for yourself, play with the parts, and discover what you can build.

With that in mind, we started a project called Au-to-do: a full sample application implementing a ticket tracker, built using Google APIs, that developers can download and dissect.


Au-to-do currently uses the following APIs and technologies:

• Google App Engine (with the Python runtime, Datastore API, and Task Queues API)
• Google Cloud Storage
• Google Prediction API
• Google Tasks API
• OAuth 2.0

Additional integrations with Google APIs are on their way. We are also planning a series of follow-up blog posts discussing each of the integrations in depth, with details on our design decisions and best practices you can use in your own projects.

By the way, if you’re wondering how to pronounce Au-to-do, you can say "auto-do" or "ought-to-do" — either is correct.

Ready to take a look at the code? Check out the getting started guide. Found a bug? Have a great idea for a feature or API integration? Let us know by filing a request.

Happy hacking!


Dan Holevoet joined the Google Developer Relations team in 2007. When not playing Starcraft, he works on Google Apps, with a focus on the Calendar and Contacts APIs. He's previously worked on iGoogle, OpenSocial, Gmail contextual gadgets, and the Google Apps Marketplace.

Posted by Scott Knaster, Editor

Author Photo
By Zachary Goldberg, Product Manager

Since the general availability launch of the Prediction API this year at Google I/O, we have been working hard to give every developer access to machine learning in the cloud to build smarter apps. We’ve also been working on adding new features, accuracy improvements, and feedback capability to the API. Today we take another step by announcing Prediction v1.4. With the launch of this version, Prediction is graduating from Google Code Labs, reflecting Google’s commitment to the API’s development and stability. Version 1.4 also includes two new features:

• Data Anomaly Analysis
• One of the hardest parts of building an accurate predictive model is gathering and curating a high quality data set. With Prediction v1.4, we are providing a feature to help you identify problems with your data that we notice during the training process. This feedback makes it easier to build accurate predictive models with proper data.
• PMML Import
• PMML has become the de facto industry standard for transmitting predictive models and model data between systems. As of v1.4, the Google Prediction API can programmatically accept your PMML for data transformations and preprocessing.
• The PMML spec is vast and covers many, many features. You can find more details about the specific features that the Google Prediction API supports here.

[フレーム]

We’re looking forward to seeing what you create with these new capabilities!

Feel free to find us and ask questions about these new features on our discussion group or submit feedback via our feedback form.


Zachary Goldberg is Product Manager for the Google Prediction API. He has a strange fascination with the Higgs Boson.

Posted by Scott Knaster, Editor

By Travis Green, Product Manager

Over the last year, the Prediction API has given you more and more tools to make your apps smarter and teach them to adapt and learn. Today we're adding a frequently requested feature: the ability to adjust models to get better performance.

Historically, getting the right predictive model has required detailed knowledge of algorithmic behavior and experience with similar datasets, and a lot of guess-and-check. With the Prediction API, we ask you what behavior you want to see, and search across many algorithms to find the best-matching one.

How it works:

1. Upload data to Google Storage for Developers.
2. Ask the Prediction API to find a great predictive model.
3. [new] Examine more detailed statistics about your model’s performance, including more training metadata and better accuracy statistics through a confusion matrix.
4. Improve performance.
1. Give your model more samples to learn from.
2. Add in more information (see these samples).
3. [new] Show the API what data is most important (categorical data only).

For those of you ready to get started, feel free to jump in through our newly updated code samples.

Travis Green's favorite part about his job is designing smart applications. In his spare time, he is in the great outdoors (looking for trouble).

Posted by Scott Knaster, Editor

By Travis Green, Product Manager

Now your apps can get smarter with as little as a single line of code. They can learn to continually adapt to changing conditions and to integrate new information. This week at Google I/O, we’re making the Google Prediction API generally available, meaning you can create apps with these capabilities for yourself. Additionally, we’re introducing several significant new features, including:

• The ability to stream data and tune your predictive models
• A forthcoming gallery of user-developed, pre-built models to add smarts even faster.

The Google Prediction API can be used by almost any app to recommend the useful, extract the essential, and automate the repetitive. For example:

• Recommend a new movie to a customer.
• Identify most important customers.
• Automatically tag posts with relevant flags.

For example, Ford Motor Co. Research is working to use the Prediction API to optimize plug-in hybrid vehicle fuel efficiency by optionally providing users with likely destinations to choose from, and soon, optimizing driving controls to conserve fuel. Because the API is a cloud-hosted RESTful service, Ford has been able to access its computationally-intensive machine learning algorithms to find patterns that rank potential destinations based on previous driving paths. Ford will be demonstrating their work at the API’s I/O Session.

Here’s a summary of the features we added to the API today:

• Streaming training data: Continually incorporate feedback for fast-adapting systems (e.g. user-chosen tags vs predicted ones, final purchases vs expected).
• General availability: Anyone can now sign up to use the API. Paid users also receive a 99.9% SLA with increased quota.
• New JavaScript library: Now deploy the Prediction API in your JavaScript – in addition to our updated Python and Java libraries.

Today, we are also announcing the Prediction API’s forthcoming gallery of pre-trained third party predictive models (try these demo models right now), and we will be adding more constantly (maybe yours – waitlist). Once complete, all Prediction API users will be able to:

• Subscribe to others’ models: improve your apps with others’ predictive data tools.
• Sell access to your models (e.g. sentiment analysis on social media).
• Import customized models through the open-standard PMML encoding.

See our recent blog post for even more ideas, and get started at the Google APIs Console.

Thanks to our community of preview developers, who have played a crucial role in helping us make the Google Prediction API simpler and more powerful since its announcement last year at I/O 2010. We are thrilled to invite all developers to join them.


Travis Green's favorite part about his job is designing smart applications. In his spare time, he is in the great outdoors (looking for trouble).

Posted by Scott Knaster, Editor

By Travis Green of the Google Prediction API Team.

If you’re looking to make your app smarter and you think machine learning is more complicated than making three API calls, then you’re reading the right blog post.

Today, we are releasing v1.2 of the Google Prediction API, which makes it even easier for preview users to build smarter apps by accessing Google’s advanced machine learning algorithms through a RESTful web service.

Some technical details of the Prediction API:

• Chooses best technique from several available machine learning algorithms.
• Supported inputs: numeric data and unstructured text.
• Outputs hundreds of discrete categories, or continuous values.
• Integrates with many platforms: Google App Engine, web and desktop apps, and command line.
• v1.2 improvements:
• Simpler interface: automatic data type detection, and score normalization.
• Paid usage tier.
• Improved usage monitoring and faster signup through the APIs Console.

Ideas to make the most of the Prediction API:

• Recommendation: What products might a user be interested in? (example)
• Filter RSS feeds, user comments, or feedback: Which posts are most relevant? Should a user comment be featured? Which feedback should we look at first? (example)
• Customize homepages: Predict what content a user would like to see and populate the page with the user’s anticipated interests.
• Sentiment analysis: Is this comment positive or negative? Does a commenter support Group A or Group B?
• Message routing: Route emails to the appropriate person based on analysis of the email contents.
• See the Prediction API website for many more!

To join the preview group, go to the APIs Console and click the Prediction API slider to “ON,” and then sign up for a Google Storage account.

We would also like to continue to thank our supportive preview users for their help making the API the service it is today. We look forward to seeing many more of you join us in making the web just a little bit smarter, and hearing your thoughts and feedback through our discussion group.

Travis Green's favorite part about his job is designing smart applications. In his spare time, he is in the great outdoors (looking for trouble).

Posted by Scott Knaster, Editor

Since its announcement at Google I/O, the Google Prediction API has seen an outstanding response from the developer community. Developers participating in the Prediction API preview are already using it to identify spam, categorize news, and more.

Today we’re adding new features to the Prediction API to make your apps even smarter:

Multi-category prediction: Imagine you’re writing a news aggregator that suggests articles based on the kinds of stories the user has read before. Previously, using the Prediction API, each article could only be tagged with one label - the most pertinent one. For example, an article about a new truck might be labeled as “truck,” but not “roomy” or “quiet.” Now articles can be tagged with all of those labels, with the labels ranked by pertinence, enabling your app to make better recommendations.

Continuous Output: You’d like to create a wine recommendation app. Matching a wine to personal preferences is a tricky task, dependent on many factors, including origin, grape, age, growing environment, and flavor presence. Previously, your app could only label wine as “good,” “decent,” “bad,” or some other set of pre-defined values. Using the new continuous output option, your app can provide a fine-grained ranking of wines based on how well they fit the user’s preferences.

Mixed Inputs: You’re creating an automatic moderator for your blog. You could already classify incoming posts automatically based on comment text and the username of the poster (text inputs), but not the number of times they’ve posted before or the number of users that have liked their posts (numeric inputs). We’ve now added support for mixed inputs, so both numeric and text data can be incorporated in your moderation helper, greatly improving accuracy and letting you get back to making content rather than managing it.

Combining Continuous Output with Mixed Inputs: To further enhance your automatic moderator, you can use continuous output to set thresholds for automatic posting, automatic rejection and manual moderation, further reducing your workload.

You can get all the details about these and other new features on the Prediction API website. We are continuing to offer the Prediction API as a preview to a limited number of developers. There is no charge for using the service during the preview. To learn more and sign up for an invitation, please join the waitlist.

By Travis Green, Prediction API Team