one does not t simply deploy ml into producti tion
play

ONE DOES NOT T SIMPLY DEPLOY ML INTO PRODUCTI TION Henrik Brink - PowerPoint PPT Presentation

ONE DOES NOT T SIMPLY DEPLOY ML INTO PRODUCTI TION Henrik Brink Machine Learning Engineering @ Wise.io / GE Digital brinkar Agenda From space to industrial machine learning Challenges Optimization dimensions Infrastructure


  1. ONE DOES NOT T SIMPLY DEPLOY ML INTO PRODUCTI TION Henrik Brink Machine Learning Engineering @ Wise.io / GE Digital brinkar

  2. Agenda • From space to industrial machine learning • Challenges • Optimization dimensions • Infrastructure • Development and deployment • Solutions • The ML meta-algorithm • Containerization • Best engineering practices • Wrap-up and questions brinkar

  3. From astronomy to industrial machine learning... brinkar

  4. brinkar

  5. brinkar

  6. brinkar

  7. brinkar

  8. brinkar

  9. brinkar

  10. brinkar

  11. Everyone ready...? brinkar

  12. ONE DOES NOT SIMPLY DEPLOY ML INTO PRODUCTION brinkar

  13. Wh What t to opti timi mize for when bui buildi ding ng ML ML systems? brinkar

  14. Accuracy? Accu cy? brinkar

  15. Implementatio Implemen ion c n cost? brinkar

  16. Ru Runti time 1920 CPUs co cost? 280 GPUs brinkar

  17. In Inter erpr pretabilit ability? brinkar

  18. LOCAL INTERPRETABLE MODEL- AGNOSTIC EXPLANATIONS (LIME arxiv.org/abs/1602.04938 brinkar

  19. Au Autom omati tion on vs vs aug augmen mentatio ion? n? brinkar

  20. Wh What t to opti timi mize for when bui buildi ding ng ML ML systems? • Accuracy? • Implementation cost? • Runtime cost? • Interpretability? • Automation vs augmentation? brinkar

  21. A A 5-st step ML meta-alg algorit ithm hm 1. Identify and define the problem 2. Collect and understand the data 3. Build and deploy a simple model that works end-to-end 4. Iterate to optimize an deploy improved models (inner loop) 5. Monitor and back-propagate changes to problem parameters (outer loop) brinkar

  22. 1. Identify and define the problem 5. Monitor and back-propagate changes to problem parameters brinkar

  23. 2. Collect and understand the data brinkar

  24. 2. Collect and understand the data brinkar

  25. 2. Collect and understand the data brinkar

  26. 3. Deploy simple model end-to-end 4. Iterate to optimize an deploy improved models Simplest possible model that solves the • problem End-to-end production deployment: • automated deployment, testing, logging, monitoring, feedback brinkar

  27. Getting cl closer... brinkar

  28. Mach chine learning infrastruct cture brinkar

  29. Continuous integration Co brinkar

  30. brinkar

  31. 3 r 3 reaso eason f n for using using c container ainers in s in mac machine lear hine learning ning... brinkar

  32. 1 P 1 Pac ackag aging ing brinkar

  33. 2 Inher 2 Inherit itanc ance tensorflow FROM tensorflow/tensorflow:latest-gpu # Your specialized modeling pipeline images text FROM my-special-pipeline # Your even more specialized pipeline segmentation classification sentiment brinkar

  34. Immutabilit Immut ability brinkar

  35. “Data should go through the exact same pipeline when making predictions, as when the model was built.” -- Good ML practitioner brinkar

  36. Us Use gr great t framewor orks and service ces... ... brinkar

  37. Op Open sou ource ce ML framewor orks brinkar

  38. RISELab • Deploy models trained in your choice of framework to Clipper with a few lines of code by using an existing model container or Clipper writing your own • Easily update or add models to running applications • Use adversarial bandit algorithms to dynamically select best model for prediction at serving time • Set latency service level objectives for reliable query latencies • Run each model in a separate Docker container for simple cluster management and resource allocation • Deploy models running on CPUs, GPUs , or both in the same application brinkar

  39. Ho Hosted M ed ML ser servic ices es brinkar

  40. Mach chine learning infrastruct cture • Continuous integration • Containerization • Utilize ML platforms brinkar

  41. Wrap up... Wr brinkar

  42. Real-World Machine Learning manning.com/brink brinkar

  43. nordic.ai meetup.com/datacph brinkar

  44. Questi Qu tions? s? brinkar

Recommend


More recommend