How to combine Db2 on Z, IBM Db2 Analytics Accelerator and IBM Machine Learning on z/OS for Credit Scoring
2 likes222 views
This document provides an overview and demonstration of combining Db2 on Z, IBM Db2 Analytics Accelerator, and IBM Machine Learning on z/OS for credit scoring applications. It discusses machine learning basics and the machine learning workflow. It then reviews how the Db2 Analytics Accelerator can be used for in-database analytics and machine learning. Finally, it demonstrates IBM Machine Learning for z/OS, including model creation, management, deployment, and continuous performance monitoring capabilities. A live demonstration of a credit scoring application that leverages these technologies is also provided.
How to combine Db2 on Z, IBM Db2 Analytics Accelerator and IBM Machine Learning on z/OS for Credit Scoring
- 1. How to combine Db2 on Z, IBM Db2 Analytics Accelerator and IBM Machine Learning on z/OS for Credit Scoring A live Demo ! March 12th - 16th , 2018 Guillaume ARNOULD arnould@fr.ibm.com Db2 Update Days 2018
- 2. © 2017 IBM Corporation IBM Confidential Agenda Analytics on IBM z Systems • Machine Learning Basics • IBM Db2 Analytics Accelerator for Machine Learning • IBM Machine Learning for z/OS + Demo
- 3. © 2017 IBM Corporation IBM Confidential Machine learning is everywhere, influencing nearly everything we do… Netflix personalized movie recommendations Waze personalized driving experience 7 out of 10 financial customers would take recommendations from a robot advisor Machine Learning Basics ▪ Identifies patterns in historical data ▪ Builds/trains behavioral models from patterns ▪ Makes recommendations
- 4. © 2017 IBM Corporation IBM Confidential The Machine Learning Workflow: Perception
- 5. © 2017 IBM Corporation IBM Confidential Machine Learning 101: Types of machine learning • Classification – Data points are labeled and are being used to predict a category – Two-class vs multi-class – Example: • Fraud detection (fraud vs non-fraud) • Spam email detection (spam vs non-spam) • Regression – When a value is being predicted – Example: • Stock prices prediction • Clustering – Data points are not labeled. – Goal is to group data into clusters to better organize the data
- 6. © 2017 IBM Corporation IBM Confidential Machine Learning 101: Supervised Learning • A feature is a piece of information that might be useful for prediction – Example, predict the probability of a customer buying a product • Labeled data is the desired output data – Example, 1.0 representing a customer has bought a product; 0.0 representing NOT GENDE R AGE MARITAL_STATUS PROFESSIO N CUSTOMER_ID LABEL F 24 Married Retail 4003 1.0 M 43 Married Trades 4004 1.0 F 43 Unspecified Hospitality 4005 0.0 F 43 Unspecified Sales 4006 1.0 M 28 Single Trades 4007 1.0 Feature Feature Feature Feature NOT a feature Label
- 7. © 2017 IBM Corporation IBM Confidential Training a model Feature Engineering Feature Engineering Scoring Labeled examples Training Scoring New data Model Model Predicted data Deploy Data Science Experience Operational system Dev Ops Machine Learning 101 : a TrainOps (DevOps) Story
- 8. © 2017 IBM Corporation IBM Confidential 8 Fraud Detection Example : a 2 Steps Approach Analyses Segments Profiles Scoring models ... Scoring Scoring based decisionData: Demographics Account activity Transactions Channel usage Service queries Renewals … Identify predictive models/patterns found in historical data Use those predictive models with variables to score transactions & identify the best possible future outcomes Practical scoring approaches ▪Off-line: Batch Scoring ▪On-line: External scoring function ▪On-line: Within a transaction, in real time Step 1 – Build the predictive model Step 2 – Execute the predictive model Model
- 9. © 2017 IBM Corporation IBM Confidential The Machine Learning Workflow: Reality
- 10. © 2017 IBM Corporation IBM Confidential Agenda Analytics on IBM z Systems • Machine Learning Basics • IBM Db2 Analytics Accelerator for Machine Learning • IBM Machine Learning for z/OS + Demo
- 11. © 2017 IBM Corporation IBM Confidential Version 5.1 of Db2 Analytics Accelerator opens up a new dimension of Analytical Processing by introducing In-Database Analytics and Accelerator-Only Tables. ➢ In-Database Analytics capabilities enable acceleration of predictive analytics applications. This enables SPSS/Netezza Analytics Data Mining and In-Database Modeling to be processed within the Accelerator. ➢ Accelerator-only tables can benefit statistics and analytics tools that use temporary data for reports. The high velocity of execution enables these tools to quickly gather all required data. ➢ Accelerator-only tables enable acceleration of Data Transformations implemented via SQL statements. Storing interim results in accelerator-only tables enables subsequent queries or data transformations to process all relevant data on the accelerator with high speed
- 12. © 2017 IBM Corporation IBM Confidential Introducing Accelerator-Only Table type in Db2 for z/OS Creation (DDL) and access remains through Db2 for z/OS in all cases Non-accelerator Db2 table • Data in Db2 only Accelerator-shadow table • Data in Db2 and the Accelerator Accelerator-archived table / partition • Empty read-only partition in Db2 • Partition data is in Accelerator only Accelerator-only table (AOT) • “Proxy table” in Db2 • Data is in Accelerator only Table 1 Table 4 Table 3 Table 2Table 2 Table 4 Table 3
- 13. © 2017 IBM Corporation IBM Confidential Data scientist work area Using Accelerator-only tables for ad-hoc analysis Transaction Processing Systems (OLTP) Data for transactional and analytical processing Customer Transactions Customer Data Customer Transactions Customer Data Work database John Work Area AOTs Work database Bob Work Area AOTs Data Scientist John Data Scientist Bob
- 14. © 2017 IBM Corporation IBM Confidential 14 ETL on a different Platform (Traditional Approach) Database Transformation – Common Usage Customer data Customer Transactions Transaction Processing Systems (OLTP) Customer Transaction Summary and History Customer data Customer Transactions Customer Summary Mart Distributed DBMS ETL logic Copy Table Data (FTP) Disadvantages: ▪ Process driven movement of large amounts of data ▪ Aged data for analytics/reporting depending on performance of data movement and transformation process Analytics Unix Server The Secret Weapon
- 15. © 2017 IBM Corporation IBM Confidential In-Database Transformation Using Accelerator-only tables and ELT Logic in the Accelerator Transaction Processing Systems (OLTP) Analytics Advantages: • Simpler to manage • Better performance and reduced latency Data for transactional and analytical processing Customer Transactions Customer Data Customer Transaction Summary and History AOTs Customer Summary Mart AOTs Customer Transactions Customer Data ELT logic
- 16. © 2017 IBM Corporation IBM Confidential 16 ETL using Infosphere Information Server
- 17. © 2017 IBM Corporation IBM Confidential 17 Balanced Optimization Using IDAA : Optimization
- 18. © 2017 IBM Corporation IBM Confidential 18 Run-Time Comparisons – Benefits of Running inside IDAA
- 19. © 2017 IBM Corporation IBM Confidential In-Database Analytics – Technical basics ▪ Set of stored procedures of IBM Netezza In-Database Analytics Package (INZA) are available on the Accelerator, including: ▪ Decision Tree ▪ Regression Tree ▪ Naive Bayes ▪ K-means Clustering and TwoStep Clustering ▪ Stored procedures use accelerator-shadow tables or accelerator-only tables as input and create accelerator-only tables as output ▪ Db2 for z/OS contains stored procedure wrappers to enable invocation of the stored procedures from Db2 for z/OS analytical applications or from SPSS Modeler 17.1 ▪ Actual stored procedure execution takes place on Accelerator
- 20. © 2017 IBM Corporation IBM Confidential In-Database Analytics Data Preparation (using AOTs) and SPSS Modeling in the Accelerator Transaction Processing Systems (OLTP) With embedded scoring Advantages: • Allows fast model refreshes • Ensures adequate scoring • Better performance and reduced latency • Scoring outside accelerator with SPSS Modeler Server , or Zementis Data for transactional and analytical processing Customer Transactions Customer Data Customer Txn Data Prep AOTs Customer Transactions Customer Data Modeling ModelModel SPSS Modeler
- 21. © 2017 IBM Corporation IBM Confidential 21 « Behind the scenes»: Exploiting In-Database Modeling using Accelerator
- 22. © 2017 IBM Corporation IBM Confidential 22 « Behind the scenes»: Execution Results
- 23. © 2017 IBM Corporation IBM Confidential 23 « Behind the scenes»: Looking at the Accelerator using DataStudio
- 24. © 2017 IBM Corporation IBM Confidential In-Database Analytics Using Accelerator-Only Tables , ELT Logic and Modeling in the Accelerator Transaction Processing Systems (OLTP) Data for Transactional and Analytical Processing Customer Transactions Customer Data Customer Transactions Customer Data Customer Transaction Summary and History AOTs Customer Summary Mart AOTs ELT logicCustomer Txn Data Prep AOTs Modeling ModelModel SPSS Modeler
- 25. © 2017 IBM Corporation IBM Confidential Agenda Analytics on IBM z Systems • Machine Learning Basics • IBM Db2 Analytics Accelerator for Machine Learning • IBM Machine Learning for z/OS + Demo
- 26. © 2017 IBM Corporation IBM Confidential 26 Understand the 2 Different Steps Decision Management / Rules application Scoring against an existing model 5 4 data synchronisation IBMDb2Analytics Accelerator TRANSACTION Rule & Model Execution Rule & Model Creation In Database Transformation 2a (accelerated) Real Time / Predictive Analytics / Reporting 2 3 Merging non Db2 for zOS data and Db2 data 1 Model & Rule Updates zDatazApps
- 27. © 2017 IBM Corporation IBM Confidential IBM ML Training runtimes Leverage a wide range of platforms to meet the varying needs of enterprise architectures IBM ML authoring (DSX) (Cloud, MacOSX, Windows, Linux, Linux on System z) Cloud Distributed Power z/OS Model Repository Spark Anaconda Deep Learning IBM Machine Learning
- 28. © 2017 IBM Corporation IBM Confidential IBM ML Deployments Leverage a wide range of platforms to meet the varying needs of enterprise architectures IBM ML authoring (DSX) (Cloud, MacOSX, Windows, Linux, Linux on System z) Cloud Distributed Power z/OS Model Repository IBM Machine Learning
- 29. © 2017 IBM Corporation IBM Confidential • Auto-modeling – Cognitive assistant for data scientists (CADS) • Select the best algorithm with the best performance from a set of candidates – Hyperparameter optimization (HPO) • Select the hyperparameter with the best performance from a set of candidates given a specific algorithm – CADS and HPO use the performance of models on small data sets to predict performance on large data sets. They use machine learning to facility machine learning. • Visualization – Data Scientists use visualization tool to help them understand data distribution. Brunel is one of the tools commonly used by Data Scientists – Brunel is designed as a layer on top of low-level visualization technology that does not require programming to design compelling interactive visualizations. Model Creation – Build-in Libraries %%brunel data(‘churndata’) bar x(AGE) y(#count) label(NEGTWEETS) color(CHURN_LABEL)
- 30. © 2017 IBM Corporation IBM Confidential • The Jupyter Notebook is an open-source web application that allows you to create and share documents that contain live code, equations, visualizations and explanatory text. Model Creation – Integrated Jupyter Notebook Cell for code snippet Interactively execution
- 31. © 2017 IBM Corporation IBM Confidential • Visual model builder is a wizard guiding users to create a model step by step • No programming skill is required Model Training – Visual Model Builder
- 32. © 2017 IBM Corporation IBM Confidential • The Predictive Model Markup Language (PMML) is an XML-based predictive model interchange format. • Many vendors can export their models to PMML format, including SPSS, R and SAS • IBM Machine Learning for z/OS supports scoring for PMML models that conforms PMML standard • Support for PMML extensions is not guaranteed PMML Model Support
- 33. © 2017 IBM Corporation IBM Confidential • Models are managed in a central repository in Db2 for z/OS • Leverage the high availability of Db2 and z Model Management – Saving Model Db2 for z/OS V10 or above Models and metadata of models ML libraries / services to persistent models Notebook Visual model builder ML services PMML Model
- 34. © 2017 IBM Corporation IBM Confidential • Model deployment is the process of moving model into production environment to serve business need – single click deployment • Models are deployed as REST interfaces • Runtime performance monitoring for scoring services Model Deployment CICS WAS Mobile DFHJSON
- 35. © 2017 IBM Corporation IBM Confidential Continuous Performance Monitoring Day 1 Day 2 Day 3 Day 4 Day 5 … New inserted feedback data Feedback Dataset GENDER AGE MARITIAL _STATUS PROFESSION … GENDER AGE MARITIAL _STATUS PROFESSION INSERT_TIME Evaluated feedback data Training Dataset Evaluate Evaluate Evaluate Evaluate Evaluate Users add new labeled data as feedback data to feedback dataset Evaluation tasks are scheduled to monitor performance of a model with the new inserted feedback data Re-train?
- 36. © 2017 IBM Corporation IBM Confidential Highlights deployments whose performance is downgrading Continuous Performance Monitoring (cont.)
- 37. © 2017 IBM Corporation IBM Confidential Components for a Machine Learning Implementation 38 Build / Train Model Evaluate Scoring Service Monitor Success Business Process / Application Historical Data New Data Training / Learning Application Scoring Ingest Deploy Feedback
- 38. © 2017 IBM Corporation IBM Confidential IBM Machine Learning for z/OS Architecture ➢ Move Machine Leaning capability to the platform where the most valuable data resides ➢ Integrate real-time predictive analytics with transactions ➢ Leverage z/OS superior reliability, availability and security
- 39. © 2017 IBM Corporation IBM Confidential The former Loan Application 40 • A customer is eligible for a loan according to several criteria such as the amount of the loan, the yearly income of the borrower, and the duration of the loan. • The decision logic is embedded in multiple loan approval applications • The branch application running on z/OS (Cobol/CICS) • The Internet application running on JEE server Branch Application (Cobol/CICS) Decision logic Internet Application (JEE) validation eligibilityvalidation eligibility Batch Scoring Application scoring ▪ Scoring is not real time ▪ Disruption in the predictive models lifecycle management (Data scientist & IT) ▪ Rules written in software code cannot be read by business people ▪ Hard coded rules are difficult to change ▪ Rules intertwined within applications cannot be reused by other systems
- 40. © 2017 IBM Corporation IBM Confidential The NEW Loan Application 41 • Objectives • The SBSLoan application illustrates how Decision Management helps lenders to make an online decision for loans approval. • Functional Description • Manage loan approval through • a set of Business Rules • An In-Transaction, In-Database real-time Scoring Real-time scoring validation eligibility scoring
- 41. © 2017 IBM Corporation IBM Confidential The NEW Loan Application – Demonstration overall picture 42 z/OS Rule Execution Server on WAS for z/OS zRES on CICS validation eligibility Decision Center Repository Business User Database Server Db2 z/OS Production Data Historical Data Scoring Services Liberty on z/OS Scoring REST/JSON Machine Learning / Spark on z/OS IBM Machine Learning UI Jupyter Notebook / Visual Model Builder / Model Management / Model Deployment / Monitoring Ingestion lib Training lib Db2 JDBC driver CADS/HPO lib validation eligibility CICS Cobol App Runtime Deployment Development / Training
- 42. © 2017 IBM Corporation IBM Confidential Live Demonstration
- 43. © 2017 IBM Corporation IBM Confidential
- 44. © 2017 IBM Corporation IBM Confidential (Relatively) New IBM Redbooks publication: SG24-8314-00 • Analytics on z Systems environment • Warehouse concepts • Logical data warehouse • Transformation patterns • Accelerator-only tables • Concepts and architecture • Use cases enabled by accelerator-only tables and in-database-analytics • Multi-step reporting • Using QMF to store query results and importing tables • Accelerating IBM Campaign processing • In-database transformations • Accelerator and accelerator-only table usage within DataStage • Accelerator-only tables supporting data scientists ad-hoc analysis • Integration of more data sources and archiving for analytics • In-database analyticshttp://www.redbooks.ibm.com/Redbooks.nsf/RedbookAbstracts/sg248314.html?Open
- 45. © 2017 IBM Corporation IBM Confidential New IBM Redbooks publication to come very soon …. • Analytics on z Systems environment • Warehouse concepts • Logical data warehouse • Transformation patterns • Accelerator-only tables • Concepts and architecture • Use cases enabled by accelerator-only tables and in-database-analytics • Multi-step reporting • Using QMF to store query results and importing tables • Accelerating IBM Campaign processing • In-database transformations • Accelerator and accelerator-only table usage within DataStage • Accelerator-only tables supporting data scientists ad-hoc analysis • Integration of more data sources and archiving for analytics • In-database analyticshttp://www.redbooks.ibm.com/Redbooks.nsf/RedbookAbstracts/sg248314.html?Open IBM Machine Learning on z/OS
- 46. © 2017 IBM Corporation IBM Confidential Proof of Concept on IBM Machine Learning for z/OS Help testing IBM Machine Learning on a Real environment. 47 Preparation Enablement Setup • How can we benefit from IBM Machine Learning solution ? • How to identify best use cases to get started with ? • How to implement IBM Machine Learning within our environment ? • How to ingest / Model our data within Machine Learning ? • How to deploy new models to be used in our current applications ? • How to manage IBM Machine Learning installation ? • See immediate benefits from IBM Machine Learning in identifying new patterns within their data. • Get help to identify, implement, ingest, train, model the most relevant data for their use cases. • Experiment in a dedicated PoC environment IBM Machine Learning solution. • Assistance to install IBM Machine Learning solution in their existing production environment. • Customer team members enabled to leverage IBM Machine Learning benefits. 2 days 1 day 3 days to 2 weeks • Client Data Scientists • Client Database Administrators • Client Technical Team • IBM Local Team Questions our clients ask themselves The benefits for our clients Participants Duration Key Activities ▪ 1 Questionnaire to be filled in by client ▪ 1H Conference call. ▪ Review of pre- requisites in client target environment • Teaching on new IBM Machine Learning for z/OS benefits and features. • Hands-On Labs on IBM Sandboxes Contacts • Preparation of IBM Machine Learning environment in Montpellier PoC dedicated environment. • Implement IBM Machine Learning environment by customer location. Guillaume Arnould Analytics on zSystems Certified IT Specialist IBM Client Center Montpellier arnould@fr.ibm.com PoC 1 month • Remote assistance while PoC .
- 47. © 2017 IBM Corporation IBM Confidential Thank You for attending the Db2 Update Days 2018 !!!