How we could abstracting the ML implementation including using keras

1. ML Abstraciton for Keras to
Serve Several Cases
Hendri Karisma

2. Actual Topic :
Micro Kernel Architecture
Implementation for Machine Learning
Hendri Karisma

3. Hendri Karisma
● Principal R&D and Data Lead at Akar Inti Data
● Working for advance data analytics platform, big data, geographic performance
● Working with Software Engineering team for Data marketplace (https://nusadata.ai )
● Data solution architect

5. What is 3rd Party Data

6. Data Marketplace
Open Finance
Telco
eKYC
Client/Consumer
Client/Consumer
Nusa Data
(Market Place)
Client/Consumer
Package
Raw
Aggregate

7. Why we need to abstract the solution?
● We facing differents cases
● Different solutions
● Different approachments
● Different methods
● Different Technologies
● Differents deadline schedule…
● So we need to make the components reusable

8. Computational Thinking

9. Micro Kernel
Architecture

10. ML Taxonomies

11. Machine Learning basic things…
● Training
● Evaluating
● Predicting
● Data preprocessings

12. Engineering, why?
● SOLID Principal
● Reusability:
○ We use the algorithm in backend service and data pipeine
○ Difference technologies like programming language or platform
● Scalability
○ Could adopt several libraries or frameworks or our own implementation
○ Performance scale up and scale out
○ Run more efficient

13. Sample Case
We have 2 cases that need to be handled:
● Keras
● SKLearn
The possibilities are could have different data type or structures and also different model, and it could be
different libraries.

14. Class Diagram

15. Architecture

16. Lets see the sample

18. Any Questions ?

19. Thank you