Advanced Analytics at Scale: Deploying Data Science in the Enterprise

Transcription

1

2 # T C 1 8 Advanced Analytics at Scale: Deploying Data Science in the Enterprise Erik Polano Associate Solution Architect Tableau Erwin van Laar Team Lead, Product Consultancy Tableau

3 Session Overview

4 Agenda Bridging the Gap Implementation Scaling Out Q&A

5 Introduction

6 Who are we? MSc Applied Economic Analysis Thesis: Fraternity, sorority and stereotyping. A field experiment examining gender dynamics in referralbased hiring presented at American Economics Association Conference, San Fransisco,2016 Oracle, Financial Applications Consultant MSc Public Health: Healthcare Management, Innovation and Policy Thesis: Patient Satisfaction with Care among Patients with Hematuria MaetisArdyn (occupational health service), Management traineeship and team leader Loves Linux Loves Theatre

7 Two worlds (One Love) Why is Tableau useful for Advanced Analytics? Visually Present Results in Interactive Format Why is Advanced Analytics useful for Business Users? Make Better Business Decisions with more informed analysis Statistics/Machine Learning Expert Domain Expert Highly concerned with Validity Highly concerned with Output

8 Problem

9 Problem

10 Experiment Ideal World: run double blind controlled experiment Causal Impact (Google Data Science team) Approximate an experiment using control variables

11 Problem Business users are not trained in (advanced) statistical analysis Data Scientists have a critical framework without necessarily having access to what problems are most pressing to their business users

12 How to Solve This? External Service Connection Result Quicker Development of More Relevant and Informed Analysis. Improve ROI on work done by Data Scientists by making their results more visible, understood and relevant to Business Users.

13 What is the End Goal?

14 Conventional Self-Reliant Data-Driven Want Can People Want Can Experience Report Factory App Time What is the End Goal?

15 Tableau empowers the whole organization. It brings advanced analytics into the hands of people who don t necessarily have an analyst s or programmer s skill set. Alexs Thompson, Ph.D. Data Scientist, Hallmark Cards

16 Bridging the Gap

17 Visual Analytics v. Advanced Analytics Visual Analytics ( vizh(oo)əl,anə'lidiks) Data access, discovery, exploration, and information-sharing elevated by visual interactivity. Advanced Analytics (əd' vanst,anə'lidiks) Smart, automated, or otherwise advanced data access, discovery, exploration, and information-sharing, meant to push the boundaries of traditional analytics.

18 Key Principles Solutions should have four key characteristics: Easy to use Fast Powerful Visual

19 What do you need? Tableau Desktop Tableau Server External Services Rserve TabPy MatLab Tableau Desktop Tableau Server

20 How could the setup look? Tableau Desktop Tableau Server End users

21 Implementation

22 Implementation Framework Discovery Prototyping Foundation Building Scaling Out

23 Assessing business question/value (Discovery) Establish champions from all aligned departments Create a Centre of Excellence Develop and promote best practices Provide support Promote culture of (advanced) analytics

24 Prototyping Model Creation Data Discovery Starts. Build Presentation of Results through Tableau Desktop Calculation types SCRIPT_REAL, SCRIPT_STR, SCRIPT_INT, SCRIPT_BOOL Calculations are handled as Table Calculations

25 Connect to Tableau Server

26 Connect to Tableau Server tsm configuration set -- key vizqlserver.extsvc.host --value <IP address> tsm configuration set -- key vizqlserver.extsvc.port --value <Port> tsm apply-pending changes

27 Dashboard Design (Foundation Building) Use of titles/tooltips Guided Analytics Interactivity Parameters Selecting the required variables Selecting the needed statistical significance

28 Dashboard Design (Foundation Building) Visual cues (design principles) Robustness of model Can you trust the calculation? Use of triggers/shapes/colors

29 Scaling Out

30 Scaling out Two Types: Iterative processes across the organization (Number of people involved in building dashboard). Availability of Advanced Analytics Dashboards (Number of people consuming the dashboard).

31 Load Balancing HAProxy Free, Open Source application for load balancing Tableau Server HAProxy Very useful for scaling out Rserve instances Easy to Configure

32 Pre-Configuring / Performance Heating Up External Environments Define How Rserve Should Start This can include: Preloading Libraries Create Environments Define Timeouts

33 Can I use both R & Python? Flask Routing requests to either environment

34 Governance Best in Class security and permissions through Tableau Server Users Groups Projects Permissions

35 Continuous Development Foster the Centre of Excellence Set up fun activities Organize hackathons Makeover Monday Set up a Tableau Doctor hour

36 Questions?

37 Summary

38 Summary By scripting directly into Tableau, you re able to: Perform sophisticated statistical analysis Empower end users to set their own parameters and resulting scripts Easily and immediately communicate visual results Use lessons learned Don t let perfect be the enemy of the good

39 Summary Remember the four key characteristics Easy to Use, Fast, Powerful & Visual Use the framework Discovery, Prototyping, Foundation Building & Scaling Out Get your champions involved Create an iterative environment, promote feedback

40 Conclusion Just get out there, and start unlocking more possibilities!

41 R E L AT E D S E S S I O N S R you ready? Jedi stats with R & Tableau Thur 10:45am 1:15pm MCCNO L2 260 Tableau + Python = Thur 10:45pm 1:15pm MCCNO - L2-217

42 R E L AT E D S E S S I O N S Embedding Tableau for self-service data science Thur 2:15pm 3:15pm MCCNO L2 238 Data science applications with TabPy/R Wed 12:00pm 1:00pm MCCNO L2 New Orleans Theater B

43 Please complete the session survey from the Session Details screen in your TC18 app

44 #TC18 Thank you!

45 Resources - Whitepapers Advanced Analytics with Tableau: Define Analytics: The changing role of BI s favorite catch-all term Using R and Tableau

46 Resources - Framework Whitepaper: The Drive Methodology Whitepaper: The Tableau Drive Manual

47 Resources Causal Impact Brodersen, K.H., GALLUSSER, F., KOEHLER, J., REMY, N., & SCOTT, S.L. (2015). Inferring causal impact using Bayesian structural time-series models. Annals of Applied Statistics, vol. 9, pp Blog Github Paper

48