Upgrade to Pro — share decks privately, control downloads, hide ads and more …

ML on GCP: Design-Develop-Deploy

ML on GCP: Design-Develop-Deploy

The Machine Learning models developed once with limited resources has to reach multiple clients which should be capable for scaling and learning in real-time. Google's AI Platform makes it easy to take your ML projects from ideation to production and deployment, quickly and cost-effectively.
Learn about ML APIs, AutoML, and Cloud ML Engine products of Google Cloud Platform.

Charmi Chokshi

December 07, 2019
Tweet

More Decks by Charmi Chokshi

Other Decks in Technology

Transcript

  1. Hello World!

    View full-size slide

  2. How’s the JOOOSHHH?

    View full-size slide

  3. How’s the JOOOSHHH?

    View full-size slide

  4. ML on GCP:
    Design-Develop-Deploy
    by @CharmiChokshi
    Machine Learning Engineer @Shipmnts.com

    View full-size slide

  5. 90% have failed to solve this...

    View full-size slide

  6. 99.9% have failed to solve this...

    View full-size slide

  7. What we just did?

    View full-size slide

  8. What we just did?
    We solved logical and complex* math puzzles in a fraction of time

    View full-size slide

  9. Think about, How we
    solved them?

    View full-size slide

  10. Think about, How we
    solved them?
    We had inputs
    We also had outputs

    View full-size slide

  11. Think about, How we
    solved them?
    We created Rules and Patterns in our brains!!

    View full-size slide

  12. Can machines create rules
    on their own?

    View full-size slide

  13. Can machines create rules
    on their own?
    No, if it’s Classical Programming...

    View full-size slide

  14. Can machines create rules
    on their own?
    YES, if it’s Machine Learning!!!

    View full-size slide

  15. Classical Programming vs Machine Learning
    Classical
    Programming
    Machine
    Learning
    Rules
    Rules
    Data
    Data
    Answers
    Answers

    View full-size slide

  16. ML Use Cases

    View full-size slide

  17. Why so little ML Apps out there?
    ● Building and Scaling ML infrastructure is hard
    ● Operating production ML system is time consuming and
    Expensive

    View full-size slide

  18. What if, you can get
    ● Fully managed service
    ● Training using custom tensorflow graph for any ML use cases
    ● Training at scale to shorten Dev Cycle
    ● Automatically maximize predictive accuracy with HyperTune
    ● Integrated Datalab experience

    View full-size slide

  19. What if, you can get
    ● Fully managed service
    ● Taring using custom tensorflow graph for any ML use cases
    ● Training at scale to shorten Dev Cycle
    ● Automatically maximize predictive accuracy with HyperTune
    ● Integrated Datalab experience
    At a single place!!!

    View full-size slide

  20. What if, you can get
    ● Fully managed service
    ● Taring using custom tensorflow graph for any ML use cases
    ● Training at scale to shorten Dev Cycle
    ● Automatically maximize predictive accuracy with HyperTune
    ● Integrated Datalab experience
    At a single place!!!

    View full-size slide

  21. ML on Cloud Platform

    View full-size slide

  22. ML on Cloud Platform

    View full-size slide

  23. ML on Cloud Platform

    View full-size slide

  24. ML on Cloud Platform

    View full-size slide

  25. ML on Cloud Platform
    DIY

    View full-size slide

  26. Machine Learning
    as an API
    Access pre-trained models with a single REST API request

    View full-size slide

  27. Available Pre-trained APIs

    View full-size slide

  28. Available Pre-trained APIs

    View full-size slide

  29. Available Pre-trained APIs

    View full-size slide

  30. Let’s identify series characters!

    View full-size slide

  31. Let’s try the Vision API!

    View full-size slide

  32. What if you want to train these
    APIs on custom data?

    View full-size slide

  33. Cloud AutoML
    Train custom models without writing a single line of code

    View full-size slide

  34. The Aha! moment

    View full-size slide

  35. How it works?

    View full-size slide

  36. जीसीपी ही तसरा बाप हे, अहम् ब्रह्माि म!

    View full-size slide

  37. What if you want to
    train/deploy/scale your Keras based
    ML model on GCP?
    Custom task having custom model and data

    View full-size slide

  38. What if your model has
    multiple inputs text and image?
    Can we get predictions for such cases using AutoML?

    View full-size slide

  39. Cloud ML Engine
    Build, Train, and Serve custom models with your own data

    View full-size slide

  40. Ways to create custom models on GCP

    View full-size slide

  41. Some of the GCP offerings

    View full-size slide

  42. Cloud ML Engine: IaaS

    View full-size slide

  43. Where ML Engine fits in ML workflow

    View full-size slide

  44. Components of AI Platform
    ● Training service
    ● Prediction service
    ● Notebooks
    ● Data labeling service (beta)
    ○ Submit a request to label your video, image, or text data with some instructions
    ● Deep learning VM image
    ● Tools to interact with AI Platform
    ○ Google Cloud Console
    ■ Stackdriver Logging
    ■ Stackdriver Monitoring
    ○ REST API

    View full-size slide

  45. Design-Develop-Deploy

    View full-size slide

  46. 3 things
    from this session...

    View full-size slide

  47. Cloud ML Engine

    View full-size slide

  48. Learn more...

    View full-size slide

  49. Questions?
    Comments?
    Suggestions?
    Machine Learning on GCP: Design-Develop-Deploy
    by @CharmiChokshi

    View full-size slide

  50. Thank you!
    Happy Learning :)
    Machine Learning on GCP: Design-Develop-Deploy
    by @CharmiChokshi

    View full-size slide