Putting Data Science Models into production

An exploration of how to use ScienceOps to get a data science model into production. Key words: Data Products, Data Science, Mathematical Modelling, Ordinary Differential Equations

Transcript

1. peadarcoyle@googlemail.com All opinions my own Data Products Data Products Or how to get models into production Data Science Meetup Luxembourg October 29th, 2014
2. Who am I? Who am I? I work as a Data Analytics Consultant for an IT Consultancy. Masters in Mathematics Specialized in Statistics and Machine Learning Recently I've been an analytics product architect
3. When I was a data science neophyte I When I was a data science neophyte I ran into this conversation... ran into this conversation...
4. But then I do laugh at things like this But then I do laugh at things like this
5. And probably looked liked this And probably looked liked this
6. Well Amazon have a celebrated recommendation engine... This is what I call a 'data product' So how do analysts serve the So how do analysts serve the business? business?
7. Analytics Project Lifecycle Analytics Project Lifecycle
8. To help the business most To help the business most I believe that data science offers the most value when the models are in production. Some of us call this a 'Data Product' In this talk I will explain how to use ScienceOps from Yhat to build a model in production Why should Amazon or Google get all the fun? Or competitive advantage?
9. The story goes a little like this The story goes a little like this (Stolen from yhat slides)
11. We all get this kind of conversation We all get this kind of conversation
12. It is hard to incorporate It is hard to incorporate data into day to data into day to day operations. day operations.
13. Hiring data scientists is hard... Hiring data scientists is hard...
15. Why? Why? Well as I discovered - Software developers especially web designers speak a different language.... Try explaining what an ODE is to a web developer.... It is like explaining CSS and Javascript to me :) I'm sure there is a book that could be written about this.... Any volunteers?
16. Write models in Ruby --> Turned out ruby doesn't have an ODE solver Possible Solutions (and their Possible Solutions (and their problems) problems) Port code to Java -----> Cross language validation PMML ----> Doesn't have great language support Batch Jobs -------> High maintenance and config More tools, more work, more time More tools, more work, more time
17. So what can we do? So what can we do? Teach Software Developers math.....
18. Or more realistically Or more realistically Use a product like ScienceOps from Yhat Key Tenets 1. Work with the tools you already know 2. Iterate quickly 3. Low touch 4. No rewriting code
19. In my use case the ML server was actually running an Ordinary Differential Equation. This is specific to ScienceOps by Yhat, undoubtedly there are other products on the market. An alternative would be to build your own server and expose it via a service. The schema I used The schema I used
20. Example Example
21. Wake Vortex Project Wake Vortex Project (This is what I've been working on for a while)
22. Result looks like this... Result looks like this...
23. Who are Y-hat? Who are Y-hat? Yhat (pronounced y-hat) is a data science technology company that provides tools and systems that allow enterprises to turn data insights into data-driven products. ScienceOps, Yhat's flagship product, is a data science operations system for managing predictive and advanced decision-making APIs and workflows. From product recommendation systems to credit scoring models and customer attrition estimators, ScienceOps lets data science teams go from insight to prototype to data-driven product efficiently and at scale. (They helped me during my project so I promised to plug them)
24. So if you wanna build a data product So if you wanna build a data product
25. Why is it so hard? Why is it so hard? Different set of tools between analysts and software developers With advanced mathematics such as ODE or statistics not all languages have the libraries to do that effectively. For example Ruby doesn't have good Stats libraries Producing tools is harder than producing a report but provides a lot more value.
26. Plus Explaining this to a Software Plus Explaining this to a Software Developer is hard! Developer is hard!
27. This conversation happened This conversation happened
28. The backlash against 'big data' The backlash against 'big data' 1) There is a bit of backlash against 'big data' and data science 2) A possible solution is producing results and data products are a good way to do that.... 3) Producing results allows you to give knowledge to customers. The goal of data science is to turn data into knowledge!
29. Lessons learned Lessons learned - I can write a model in Python and have it deployed! - Software Engineers aren't data scientists and shouldn't be expected to write models in code. - Models only provide value when they are in production - Getting information from stakeholders is really valuable in improving models. - A data scientist is often a 'translator' between business and developers.
30. Challenges Challenges Data Quality is a challenge for all data science projects, this is no exception. ScienceOps has a great free service but it is a challenge when they shut you down :( Products such as ScienceOps bring us a step closer in covering the gap in understanding between Mathematician (or data scientist) and Web Developer. Someone should write a book titled 'Probability Distributions for Software Engineers'
31. Successes Successes In a few months it was possible to have an analytics product in production, using information consumed from a variety of API's. I have no idea how else - maybe using PMML that I could deploy models.
32. But when you finish a data product But when you finish a data product
33. Other kinds of data science Products Other kinds of data science Products Credit risk modelling Customer attrition modelling Recommendation engines Airline delay analysis The list goes on....
34. You can email me if you wanna chat about data peadarcoyle@googlemail.com Wanna learn more? Wanna learn more? For Yhat: www.yhathq.com They have a really good blog :)
35. Questions? Questions?