In-Database Analytics Deep Dive with Teradata and Revolution
•
4 likes•4,151 views
Teradata and Revolution Analytics worked together to develop in-database analytical capabilities for Teradata Database. Teradata v14.10 provides a foundation for in-database analytics in Teradata. Revolution Analytics has ported its Revolution R Enterprise (RRE) Version 7.1 to use the in-database capabilities of version 14.10. With RRE inside Teradata, users can run fully parallelized algorithms in each node of the Teradata appliance to achieve performance and data scale heretofore unavailable. We'll get past the market-ecture quickly and dive into a “how it really works” presentation, review implications for system configuration and administration, and then take questions from Teradata users who will be charged with deploying and administering Teradata systems as platforms for big data analytics inside the database engine.
Report
Share
In-Database Analytics Deep Dive with Teradata and Revolution
- 1. Mario Inchiosa Chief Scientist, Revolution Analytics In-Database Analytics Deep Dive with Teradata and Revolution R Tim Miller Partner Integration Lab, Teradata
- 2. • Introduction • Revolution R Enterprise • Case Study – Global Internet Marketplace • Under the Hood • Summary & Questions Agenda
- 3. • What data storage/management software do you use? > Hadoop > Teradata > LSF Clusters/Grids > Servers Please choose all that apply Poll Question #1
- 4. • Most powerful statistical programming language – Flexible, extensible and comprehensive for productivity • Most widely used data analysis software – Used by 2M+ data scientists, statisticians and analysts • Create beautiful and unique data visualizations – As seen in New York Times, Twitter and Flowing Data • Thriving open-source community – Leading edge of analytics research • Fills the talent gap – New graduates prefer R What is R? R is Hot bit.ly/r-is- hot WHITE PAPER
- 5. Exploding growth and demand for R • R is the highest paid IT skill > Dice.com, Jan 2014 • R most-used data science language after SQL > O’Reilly, Jan 2014 • R is used by 70% of data miners > Rexer, Sep 2013 • R is #15 of all programming languages > RedMonk, Jan 2014 • R growing faster than any other data science language > KDnuggets, Aug 2013 • More than 2 million users worldwide R Usage Growth Rexer Data Miner Survey, 2007-2013 70% of data miners report using R R is the first choice of more data miners than any other software Source: www.rexeranalytics.com
- 6. Debt<10% of Income Debt=0% Good Credit Risks Bad Credit Risks Good Credit Risks Yes YesYes NO NONO Income>$40K SQL Request Sample Data Debt<10% of Income Debt=0% Good Credit Risks Bad Credit Risks Good Credit Risks Yes YesYes NO NONO Income>$40K Results Desktop and Server Analytic Architecture In-Database Analytic Architecture Results Server Based vs. In-Database Architectures Why Is Teradata Different? Exponential Performance Improvement Analyst
- 7. Node level calculation: 1 2 7 9 = 4.5 • R is distributed across nodes or servers • Runs independently of the other nodes/servers > Great for row independent processing such as Model Scoring > However, for analytic functions requiring all the data such as Model Building… – Onus is on the R programmer to understand data parallelism Challenges Running R in Parallel 1 1 1 1 2 9 1 7 9 3 9 9 System level calculation: 1 1 1 1 1 2 3 7 9 9 9 9 = 2.5 Example: Median (Midpoint) Node Level 1. Find median per node 2. Consolidate and find the midpoint of the results 3. Produce the wrong answer System Level 1. Sort all the data 2. Take midpoint 3. Produce the right answer < Wrong < Right
- 8. R Operations on Data R operates on independent rows > Score models for a given observation > Parsing Text field > Log(x) R operates on independent partitions > Fit a model to a partition such as region, time, product or store R operates on the entire data set > Global sales average > Regression on all customers R Client R Client R Client
- 9. • What statistical programming tools do you use? > R/RRE > SAS > SPSS > Statistica > KXEN Please choose all that apply Poll Question #2
- 10. Who is Revolution Analytics? Revolution Analytics
- 11. OUR COMPANY The leading provider of advanced analytics software and services based on open source R, since 2007 OUR SOFTWARE The only Big Data, Big Analytics software platform based on the data science language R SOME KUDOS Visionary Gartner Magic Quadrant for Advanced Analytics Platforms, 2014
- 12. Finance Insurance Healthcare & Pharma Digital Economy Analytics Service Providers Manufacturing & High Tech
- 13. Revolution R Enterprise is…. the only big data big analytics platform based on open source R, the de facto statistical computing language for modern analytics • High Performance, Scalable Analytics • Portable Across Enterprise Platforms • Easier to Build & Deploy Analytics
- 14. Big Data In-memory bound Hybrid memory & disk scalability Operates on bigger volumes & factors Speed of Analysis Single threaded Parallel threading Shrinks analysis time Enterprise Readiness Community support Commercial support Delivers full service production support Analytic Breadth & Depth 5000+ innovative analytic packages Leverage open source packages plus Big Data ready packages Supercharges R Commercial Viability Risk of deployment of open source Commercial license Eliminate risk with open source It Has Some Limitations for Enterprises R: Open Source that Drives Innovation, but…
- 15. The Big Data Big Analytics Platform Introducing Revolution R Enterprise (RRE) DistributedR DevelopR DeployR ScaleR ConnectR • Big Data Big Analytics Ready > Enterprise readiness > High performance analytics > Multi-platform architecture > Data source integration > Development tools > Deployment tools
- 16. The Platform Step by Step: R Capabilities R+CRAN • Open source R interpreter • UPDATED R 3.1.1 • Freely-available R algorithms • Algorithms callable by RevoR • Embeddable in R scripts • 100% Compatible with existing R scripts, functions and packages RevoR • Based on open source R • Adds high-performance math Available On: • Teradata Database • Hortonworks Hadoop • Cloudera Hadoop • MapR Hadoop • IBM Platform LSF Linux • Microsoft HPC Clusters • Windows & Linux Servers • Windows & Linux Workstations
- 17. DeployR • Web services software development kit for integration analytics via Java, JavaScript or .NET APIs • Integrates R Into application infrastructures Capabilities: • Invokes R Scripts from web services calls • RESTful interface for easy integration • Works with web & mobile apps, leading BI & Visualization tools and business rules engines DevelopR • Integrated development environment for R • Visual ‘step-into’ debugger • Based on Visual Studio Isolated Shell Available on: • Windows DevelopR DeployR The Platform Step by Step: Tools & Deployment
- 18. DevelopR - Integrated Development Environment Script with type ahead and code snippets Solutions window for organizing code and data Packages installed and loaded Objects loaded in the R Environment Object details Sophisticated debugging with breakpoints , variable values etc.
- 19. DeployR - Integration with 3rd Party Software • Seamless – Bring the power of R to any web enabled application • Simple – Leverage common APIs including JS, Java, .NET • Scalable – Robustly scale user and compute workloads • Secure – Manage enterprise security with LDAP & SSO Data Analysis Business Intelligence Mobile Web Apps Cloud / SaaS R / Statistical Modeling Expert DeployR Deployment Expert
- 20. The Platform Step by Step: Parallelization & Data Sourcing ConnectR • High-speed & direct connectors Available for: • High-performance XDF • SAS, SPSS, delimited & fixed format text data files • Hadoop HDFS (text & XDF) • Teradata Database • ODBC ScaleR • Ready-to-Use high-performance big data big analytics • Fully-parallelized analytics • Data prep & data distillation • Descriptive statistics & statistical tests • Correlation & covariance matrices • Predictive Models – linear, logistic, GLM • Machine learning • Monte Carlo simulation • Tools for distributing customized algorithms across nodes DistributedR • Distributed computing framework • Delivers portability across platforms Available on: • Teradata Database • Hortonworks / Cloudera / MapR • Windows Servers / HPC Clusters • IBM Platform LSF Linux Clusters • Red Hat Linux Servers • SuSE Linux Servers
- 21. Revolution R Enterprise ScaleR: High Performance Big Data Analytics Data Prep, Distillation & Descriptive Analytics R Data Step Descriptive Statistics Statistical Tests Sampling • Data import – Delimited, Fixed, SAS, SPSS, ODBC • Variable creation & transformation using any R functions and packages • Recode variables • Factor variables • Missing value handling • Sort • Merge • Split • Aggregate by category (means, sums) • Min / Max • Mean • Median (approx.) • Quantiles (approx.) • Standard Deviation • Variance • Correlation • Covariance • Sum of Squares (cross product matrix) • Pairwise Cross tabs • Risk Ratio & Odds Ratio • Cross-Tabulation of Data • Marginal Summaries of Cross Tabulations • Chi Square Test • Kendall Rank Correlation • Fisher’s Exact Test • Student’s t-Test • Subsample (observations & variables) • Random Sampling
- 22. Revolution R Enterprise ScaleR (continued) Predictive Models • Covariance/Correlation/Sum of Squares/Cross-product Matrix • Multiple Linear Regression • Logistic Regression • Generalized Linear Models (GLM) - All exponential family distributions: binomial, Gaussian, inverse Gaussian, Poisson, Tweedie. Standard link functions including: cauchit, identity, log, logit, probit. - User defined distributions & link functions. • Classification & Regression Trees and Forests • Gradient Boosted Trees • Residuals for all models • Histogram • ROC Curves (actual data and predicted values) • Lorenz Curve • Line and Scatter Plots • Tree Visualization Data Visualization Variable Selection • Stepwise Regression • Linear • Logistic • GLM • Monte Carlo • Run open source R functions and packages across cores and nodes Cluster Analysis • K-Means Classification & Regression • Decision Trees • Decision Forests • Gradient Boosted Trees • Prediction (scoring) • PMML Export Simulation and HPC Deployment Statistical Modeling Machine Learning
- 23. DistributedR ScaleR ConnectR DeployR Write Once…Deploy Anywhere. DESIGNED FOR SCALE, PORTABILITY & PERFORMANCE In the Cloud Amazon AWS Workstations & Servers Windows Linux Clustered Systems IBM Platform LSF Microsoft HPC Hadoop Hortonworks, Cloudera, MapR EDW Teradata Database
- 24. • Challenge: Model and score 250M customers • Server-based workflow was taking 3 days • Move calculation in-database to drastically reduce runtime, process twice as many customers, and increase lift Case Study - Global Internet Marketplace
- 25. • Binomial Logistic Regression > 50+ Independent variables including categorical with indicator variables > Train from small sample (many thousands) – not a problem in and of itself > Scoring across entire corpus (many hundred millions) – slightly more challenging Existing Open Source R model
- 26. • Same Binomial Logistic Regression > 50+ Independent variables including categorical with indicator variables > Train from large sample (many millions) – more accurately captures user patterns and increases lift > Scoring across entire corpus (many hundred millions) – completes in minutes Revolution R Enterprise model
- 27. By moving the compute to the data RRE Used to Optimized the Current Process Before After Reduced 3 day process to 10 minutes
- 28. Scaling study: Time vs. Number of Rows Benchmarking the Optimized Process rows time NOTE: • Teradata Environment > 4 node, 1700 Appliance • RRE Environment > version 7.2, > R 3.0.2 Server-based (Not In-DB) In-DB
- 29. • Before trainit <- glm(as.formula(specs[[i]]), data = training.data, family='binomial', maxit=iters) fits <- predict(trainit, newdata=test.data, type='response') • After trainit <- rxGlm(as.formula(specs[[i]]), data = training.data, family='binomial', maxIterations=iters) fits <- rxPredict(trainit, newdata=test.data, type='response') Recode Open Source R to Revolution R Enterprise Optimization process
- 30. Revolution R Enterprise How RRE Scale R Actually Works
- 31. Open Source R Revolution R Enterprise Computation (4-core laptop) Open Source R Revolution R Speedup Linear Algebra1 Matrix Multiply 176 sec 9.3 sec 18x Cholesky Factorization 25.5 sec 1.3 sec 19x Linear Discriminant Analysis 189 sec 74 sec 3x General R Benchmarks2 R Benchmarks (Matrix Functions) 22 sec 3.5 sec 5x R Benchmarks (Program Control) 5.6 sec 5.4 sec Not appreciable 1. http://www.revolutionanalytics.com/why-revolution-r/benchmarks.php 2. http://r.research.att.com/benchmarks/ Customers report 3-50x performance improvements compared to Open Source R — without changing any code RevoR - Performance Enhanced R Revolution R Enterprise:
- 32. Across Cores and Nodes Scalable and Parallelized
- 33. • Anatomy of a PEMA: 1) Initialize, 2) Process Chunk, 3) Aggregate, 4) Finalize • Process a chunk of data at a time, giving linear scalability • Process an unlimited number of rows of data in a fixed amount of RAM • Independent of the “compute context” (number of cores, computers, distributed computing platform), giving portability across these dimensions • Independent of where the data is coming from, giving portability with respect to data sources “Parallel External Memory Algorithms” Scalability and Portability of PEMAs
- 34. • Efficient computational algorithms • Efficient memory management – minimize data copying and data conversion • Heavy use of C++ templates; optimal code • Efficient data file format; fast access by row and column • Models are pre-analyzed to detect and remove duplicate computations and points of failure (singularities) • Handle categorical variables efficiently ScaleR Performance
- 35. Speed and Scalability Comparison • Unique PEMAs: Parallel, external- memory algorithms • High-performance, scalable replacements for R/SAS analytic functions • Parallel/distributed processing eliminates CPU bottleneck • Data streaming eliminates memory size limitations • Works with in-memory and disk- based architectures
- 36. In-Database Billion Row Logistic Regression • 114 seconds on Teradata 2650 (6 nodes, 72 cores), including time to read data • Scales linearly with number of rows • Scales linearly with number of nodes: 3x faster than on 2 node Teradata system
- 37. Allstate compares SAS, Hadoop, and R for Big-Data Insurance Models Approach Platform Time to fit SAS 16-core Sun Server 5 hours rmr/MapReduce 10-node 80-core Hadoop Cluster > 10 hours R 250 GB Server Impossible (> 3 days) Revolution R Enterprise In-Teradata on 6-node 2650 3.3 minutes Generalized linear model, 150 million observations, 70 degrees of freedom http://blog.revolutionanalytics.com/2012/10/allstate-big-data-glm.html
- 38. • At what stage are you in your in-database analytics deployment project? > Still researching tools and methods > Evaluating/Selecting data storage/management platform > Evaluating/Selecting analytics programming tools > Launched the project/working on it now > We’re done and looking for another one! Please select one answer Poll Question #3
- 39. • Revolution R Enterprise has a new “data source”, RxTeradata (ODBC and TPT) # Change the data source if necessary tdConn <- "DRIVER=…; IP=…; DATABASE=…; UID=…; PWD=…“ teradataDS <- RxTeradata(table=“…", connectionString=tdConn, …) • Revolution R Enterprise has a new “compute context”, RxInTeradata # Change the “compute context” tdCompute <- rxInTeradata(connectionString=..., shareDir=..., remoteShareDir=..., revoPath=..., wait=.., consoleOutput=...) • Sample code for R Logistic Regression # Specify model formula and parameters rxLogit(ArrDelay>15 ~ Origin + Year + Month + DayOfWeek + UniqueCarrier + F(CRSDepTime), data=teradataDS) RRE End-User’s Perspective
- 40. • Table User Defined Functions (UDFs) allow users to place a function in the FROM clause of a SELECT statement • Table Operators extend the existing table UDF capability: > Table Operators are Object Oriented – Inputs and outputs can be arbitrary and not “fixed” as Table UDF’s require > Table Operators have a simpler row iterator interface – Interface simply produces output rows providing a more natural application development interface than Table UDF’s > Table operators operate on a stream of rows. – Rows are buffered for high-performance, eliminating row at a time processing > Table operators support PARTITON BY and ORDER BY – Allows the development of Map Reduce style operators in-database Table Operators – Teradata 14.10+
- 41. RRE Architecture in Teradata 14.10+ Worker Process Message Passing Layer Master Process … Request Response Teradata 14.10+ Data Partition Data Partition Data Partition Data Partition Master Process Worker Process Worker Process Worker Process… * All communication is done by binary BLOB’s PE Layer AMP Layer 1. RRE commands are sent to a “Master Process” - an External Stored Procedure (XSP) in the Parsing Engine that provides parallel coordination 2. RRE analytics are split into “Worker Process” tasks that run in a Table Operator (TO) on every AMP. a. HPA analytics iterate over the data, and intermediate results are analyzed and managed by the XSP. b. HPC analytics do not iterate, and final results from each AMP are returned to the XSP 3. Final combined results are assembled by the XSP and returned to the user tdConnect <- rxTeradata(<data, connection string, …>) tdCompute <- rxInTeradata(<data, server arguments, …>) ** PUT-based Installer
- 42. • High-performance, scalable, portable, fully-featured algorithms • Integration with R ecosystem • Compatibility with Big Data ecosystem Summary
- 43. PARTNERS Mobile App InfoHub Kiosks teradata-partners.com WE LOVE FEEDBACK Questions Rate this Session Questions? Resources for you (available on RevolutionAnalytics.com): • White Paper: Teradata and Revolution Analytics: For the Big Data Era, An Analytics Revolution • Webinar: Big Data Analytics with Teradata and Revolution Analytics
- 44. PARTNERS Mobile App InfoHub Kiosks teradata-partners.com WE LOVE FEEDBACK Questions Rate this Session Thank You! www.RevolutionAnalytics.com www.Teradata.com