Martin Omander from Google presented on two APIs: Google Storage for reliable scalable data storage and Google Prediction for building predictive models from historic data. He demonstrated using Storage for file hosting and Prediction for language detection. Both APIs allow leveraging Google's infrastructure at low cost, focusing on customers rather than building proprietary systems.
Bulletproof your storage and predictions with Google APIs
1. Bulletproof your storage
Predict the future
Martin Omander
momander@google.com
Developer Relations
2. Introductions
● Working for a big company (100+ employees)
● Working for a medium/small company
● Working for a startup
● Student / teacher / hobbyist
------------------------------------------------------------------
● Used the Google Storage API before
● Used / built other storage services
● Used the Google Prediction API before
● Used other / built other prediction services
3. A nice piece of equipment
Keck Observatory at Mauna Kea, Hawaii
● Adaptive optics, adjusts mirrors twice
per second to an accuracy of 4
nanometers.
● Weight: 2 x 300 tons.
● Can detect the equivalent of a candle
on the moon.
● Cost: $140 million.
● Track record: has found more
extrasolar planets than any other
observatory in the world, plus direct
evidence of the Big Bang.
4. Not everyone can own a Keck Observatory...
but not everyone has to.
The telescopes can be controlled
remotely.
In July 2011, 27 different teams of
astronomers are using the Keck
Observatory.
5. If you only remember one thing from tonight...
Focus on your customers and what you are good at.
Use someone else's telescope. Don't build your own
unless you are better at it than everyone else.
6. Tonight's agenda:
Two "telescopes" built by Google, available to you
Add reliable, scalable, fast data storage to your app:
Google Storage
Let your app predict the future based on historic data:
Google Prediction
11. How to get started
API Console:
http://code.google.com/apis/console
12. Pricing and availability
Free trial (until Dec 31, 2011)
● 5 GB of storage
● 25 GB download + 25 GB upload
● 25,000 GET, HEAD requests
Pricing
● Storage: $0.17/GB/month
● GET, HEAD requests: $0.01 per 10,000
● Upload: $0.10/GB
● Download data:
○ $0.15/GB to Americas, Europe, the Middle East, and Africa
○ $0.30/GB to Asia-Pacific
CloudHarmony: "This hybrid model of CDN-like performance with enterprise
storage features like durability, consistency and access control represents an
exciting leap forward for cloud storage!"
http://blog.cloudharmony.com/2010/06/google-storage-cdnstorage-hybrid.html
14. Google tries to predicts the future all the time
Google filters spam, detects languages, analyzes traffic
patterns, and analyzes other large data sets every day. We
have opened this up so developers can use our machine-
learning tools too.
16. How the Prediction API works, real estate demo
Upload your Use the API, gsutil or any
1. Upload training data to interoperable utility to upload
Google Storage your data to Google Storage
Build a model prediction/v1.1/training?data={}
2. Train from your data POST : a training request
GET : model info
3. Predict Make new prediction/v1.1/training/{}/predict
predictions POST : a prediction request
17. Sample Use Cases
Data Customer Content
Classification Sentiment Moderation
Product Automatic Message
Recommendation Tagging Routing
18. "Our best thoughts come from others."
-- Ralph Waldo Emerson
● Hosted model subscriptions
● Revenue sharing
● PMML import
● Streaming updates
20. A nice piece of equipment
Keck Observatory at Mauna Kea, Hawaii
● Adaptive optics, adjusts mirrors twice
per second to an accuracy of 4
nanometers.
● Weight: 2 x 300 tons.
● Can detect the equivalent of a candle
on the moon.
● Cost: $140 million.
● Track record: has found more
extrasolar planets than any other
observatory in the world, plus direct
evidence of the Big Bang.