Predicting with Google

This talk was delivered at Google DevFest by GDG Ahmedabad on 27th December, 2016. This Presentation features Google's Prediction API and what is Machine Learning as a Service on Cloud.

Transcript

1. Predicting with Google { “label”: “Awesome” “Score”: 0.99 }, { “label”: “Bad” “Score”: 0.01 }
2. What is Prediction?
3. Wait, What!? • It is ‘Machine Learning as a Service’ on Cloud. • Predicts an Output value based upon the entered input like Language, Sentiments, tags e.t.c. • Available as an on demand HTTP RESTful Web Service.
4. Okay, So What I Do Now!? • Well, pretty much the same thing that we do ‘Everyday’ PREDICT! • Starting right from churning a small file of few Kbs to a Big Data Set of 1.5GBs! you can feed them all in form of a CSV. • But! How DO I DO IT!?
5. How Does it Work? • Follow ‘3 Simple Things’ A. Upload B. Train C. Predict • And Life’s Good! Don’t waste it behind ‘Too Much Code’!
6. This Way,
7. A. • Have your date in form of a CSV file. Now remember, when I say CSV it doesn’t mean Excel! • Else, just push whatever date you have to the Cloud and make use of the Google Storage API. "english","To err is human, but to really ..." "spanish","No hay mal que por bien no venga." ... Upload to Google Storage gsutil cp ${data} gs://yourbucket/${data}
8. • Uploading is a pretty much simple process. • Use Google Cloud Storage and then fill up the details, select an App Engine Location and you are good to go!
9. B. • Create a new Model using the training data to have accurate predictions. • The Prediction API finds relevant feature from the dataset while the training process • This very features are searched by it later during the Prediction process.
10. • Above Sample Training Data shares an insight of how the Prediction API extracts the features for the Prediction purpose
11. • Google Prediction API is extracting the features for language analysis. English is Blue while Spanish is Red.
12. • While Training a model, POST prediction/v1.1/training?data=mybucket%2Fmydata • Training runs asynchronously so, to see the results, GET prediction/v1.1/training/mybucket%2Fmydata {"data":{ "data":"mybucket/mydata", "modelinfo":"estimated accuracy: 0.xx” }}
13. C. • Well, finally you are done now! • Your model is ready to Predict results with new data. • So, again, how it works!? POST prediction/v1.1/query/mybucket%2Fmydata/predict { "data":{ "input": { "text" : [ “Your Sample Text” ]}} }
14. • So, let us have a closer look at the Workflow of the process above.
15. Types of Prediction • Regression How input can cause Output to vary is seen as well the outputs obtained are Numeric in value. • Classification Need to decide which bucket an output belongs to and buckets are Text values
16. Brought to Life!
18. In a Nutshell,
19. Capabilities • Data
 Input Features: numeric or unstructured text. Output: up to 100s of discrete categories. • Training
 Many machine learning techniques. Automatically selected.
 Performed asynchronously. • Access from many environments
 Web app, mobile app, desktop app. ßProgramming language independent .
20. Usages,
21. Why Use It? • Makes Machine Learning easy to use. • Makes every App. Smarter. • Being a web service, can be accessed seamlessly from any Platform. • It also provides strategies for Large Scale Machine Learning
22. Use Case, • Consider an example of Reddit where we will guess the Sub Reddits according to the submissions. • 75MB of JSON-encoded data, comprising 72,986 submissions • This subset made up 42,753 submissions, or about 58% of the original. • Submissions were randomly split into either the training set (98%) or the validation set (2%).
23. Churned Data
24. Conclusion • Makes Life a much more ‘Simpler’! • And Yes, they too have a ‘Free Module’ in it!
25. Final Predictions • Today is going to be one of the finest day in the DevCon history of Gujarat. • Though the Cold is increasing but the temperature at DevFest will be ‘Warmer’! • Ofcourse, the ‘Food is Awesome’ !
26. EoF +ManoranjanPadhy @_manoranjanp FB: manoranjan.padhy
27. More Predictions?