Operational Machine Learning: Using Microsoft Technologies for Applied Data Science

| © Copyright 2015 Hitachi Consulting1 Operational Machine Learning Using Microsoft Technologies for Applied Data Science Khalid M. Salama, Ph.D. Business Insights & Analytics Hitachi Consulting UK We Make it Happen. Better.

| © Copyright 2015 Hitachi Consulting2 Outline  Introduction to Data Science  From Experimental Data Science to Operational Machine Learning  MS Technologies for Data Science & Advanced Analytics  Demos & Screenshots  Concluding Remarks

| © Copyright 2015 Hitachi Consulting4 Data Science and Machine Learning What? Data Science Machine Learning Statistics Artificial Intelligence Databases Other Technologies “Data mining, an interdisciplinary subfield of computer science, is the computational process of automatic discovering interesting and useful patterns in large data sets” Other Related Technologies:  Visualization  Big Data  High Performance Computing  Cloud Computing  Others..

| © Copyright 2015 Hitachi Consulting5 Data Science and Machine Learning Why? Vision Analytics Recommendation engines Advertising analysis Weather forecasting for business planning Social network analysis Legal discovery and document archiving Pricing analysis Fraud detection Churn analysis Predictive Maintenance Location-based tracking and services Personalized Insurance The objective of data science is to provide you with actionable insights to support decision making….

| © Copyright 2015 Hitachi Consulting6 Data Science and Machine Learning How? Classification Learning Build a model that can predict the target class of an input case Cluster Analysis Discover natural groupings within the data points Association Rule Discovery Extract frequent patterns present in the data Regression Modeling Build a model that can estimate the response value given an input case Time Series Analysis Analysis of temporal data to forecast future values Probabilistic Modeling Compute the probability of an event to occur given a set of conditions Similarity Analysis Identify similar cases to a given input case based on the input features Collaborative Filtering Filtering of information using techniques involving collaboration viewpoints IF .. AND .. AND .. THEN A ELSE IF .. AND .. THEN C ELSE IF .. AND .. THEN B .. .. ELSE C

| © Copyright 2015 Hitachi Consulting8 Exploratory Data Analysis Data Science Activities Experimentation vs. Operationalization Collect Data Blend Visualize Prepare ML Experiment Algorithm Selection Parameter Tuning Training & Testing Model Learning Dataset Report of Visuals & Findings Decision! Data Analysis & Experimentation  Interactive  Easy to perform  Rich Visualizations

| © Copyright 2015 Hitachi Consulting9 Online Apps Automated ML Pipeline Data Science Activities Experimentation vs. Operationalization Model Data Ingestion Data Processing Model Training Scoring Deploy Web APIs Predict Train Export Batch Real-time Operational ML Pipelines  Pipelined (ETL Integration)  Scalable  Apps Integration

| © Copyright 2015 Hitachi Consulting12 Microsoft Advanced Analytics Data Science, Machine Learning, & Intelligence Data Mining – SQL Server Analysis Services Azure Machine Learning Spark ML – Azure HDInsight Microsoft R Server – SQL Server R Services Azure Cognitive Services Cognitive Features – Azure Data Lake Analytics Microsoft Bot Framework

| © Copyright 2015 Hitachi Consulting14 Azure Machine Learning MS Cloud-native Data Science  Cloud-based Machine Learning Services  Interactive Data Science Studio  Rich built-in functionality  Imports data from everywhere  Easy to develop and productionize – Web Services  Extensible via R and Python scripts Azure Machine Learning Build and deploy models in the cloud Import Data Publish Result Input Web Services Batch Scoring Retrain Model Limitations  Only Cloud-based (Data Regulations)  Scalability – Maximum dataset size = 10GB  Microsoft R Open is not supported, yet  No Source Control

| © Copyright 2015 Hitachi Consulting15 Azure Machine Learning Real-time Predictions App Event Hub Stream Analytics Power BI Azure ML Web Service Send data points Consume messages Send Input Receive Output Send Results (Input, Output)

| © Copyright 2015 Hitachi Consulting23 Microsoft R Server R in Microsoft World Microsoft R Open (MRO)  Based on latest Open Source R (3.2.2.) - Built, tested, and distributed by Microsoft  More efficient and multi-threaded computation  Enhanced by Intel Math Kernel Library (MKL) to speed up linear algebra functions  Compatible with all R-related software

| © Copyright 2015 Hitachi Consulting24 Microsoft R Server Comparison CRAN MRO MRS Data size In-memory In-memory In-memory & disk Efficiency Single threaded Multi-threaded Multi-threaded, parallel processing 1:N servers Support Community Community Community + Commercial Functionality 7500+ innovative analytic packages 7500+ innovative analytic packages 7500+ innovative packages + commercial parallel high-speed functions Licence Open Source Open Source Commercial license.

| © Copyright 2015 Hitachi Consulting25 Microsoft R Server Components & Compute Contexts Microsoft R Server CRAN&MSROpen ScaleR DistributedR ConnectR MicrosoftML-Package Operationalization (msrdeploy) RStudio | RTVS MS R Client Scale & Deploy DifferentComputeContexts  Installed on Windows or Linux  ScaleR - Optimized for parallel execution on Big Data, to eliminate memory limitations.  ConnectR – Provides access to local file systems, hdfs, hive, sqlserver, Teradata, etc.  DistributeR - Adaptable parallel execution framework to enable running on different (distributed) compute contexts.  Operationalization (msrdeploy) – Deploy the model as a Web API.

| © Copyright 2015 Hitachi Consulting29 SQL Server R Services In-database Analytics  R Services (in-database) – Keep your analytics close to the data  T-SQL Script – Can be encapsulated in Stored Procedures  Models are built, trained, saved as part of the ETL process (SSIS)  Used for batch prediction (as part of the ETL process)  Visual Studio SQL Database Project, Source Controlled, etc.  Uses Microsoft ScaleR libraries Limitations  Not supported in Azure SQL DB/DW, yet  Not suitable for Interactive Data Science  Only R, no python, yet. Process Data Train R Model Serialize Store Models Maintain Models Process Data Load Model Perform Prediction Store Results ETL Using SSIS Data Sources Prediction Pipeline Training Pipeline EXECUTE sp_execute_external_script

| © Copyright 2015 Hitachi Consulting32 SQL Server Analysis Services Data Mining Limitations  Limited Extensibility  Limited Algorithms & Functionalities  No Azure PaaS Service Azure SQL DW/DB SQL Server Analysis Services Online Apps Build Model Result Explore/ Interpret Model DMX Query Batch Scoring Retrain Model  Process data from many OLEDB and ODBC data sources  Easy to build, interpret, deploy, and productionize  SSIS Support – Tasks to Train & Predict  Interactive Visuals for model interpretation  Excel Integration – Data Mining Add-in

| © Copyright 2015 Hitachi Consulting33 SQL Server Analysis Services Overview Data Source View Mining Structure Mining Algorithm Mining Model  Decision Tress  Naïve-Bayes  Linear Regression  Neural Networks  Association Rules  Clustering  Sequence Clustering  Time Series

| © Copyright 2015 Hitachi Consulting40 Azure Data Lake Analytics Cognitive Features  Pre-built intelligence – Text & Image Analysis  Integrated with your data processing pipelines (DLA)  Used for batch recognition (not singleton real-time)  Scheduled & Automated using Azure Data Factory  R & Python Extensions!  Scalable – Suitable for Big Data Ingest Polybase Input Output Data Processing & Patten Recognition Source Data (Text, Images, etc.) Enterprise Data Warehouse Azure SQL DW Data Lake Analytics Jobs Data Lake Store Azure Data Factory Data Lake Store Limitations  Limited Features  Not suitable for real-time scoring

| © Copyright 2015 Hitachi Consulting45 Spark ML on HDInsight Scalable ML for Big Data  Rich Spark ML Libraries  Scalable, distributed, in-memory  Extensible – Python, R, Java, Scala  Suitable for Big Data - Batch Model Training and Scoring  Spark Streaming for Real-time predictions  Scheduled & Automated Using Azure Data Factory Ingest - Process Data - Build Model - Save Model - Load Model - Perform Predictions - Save Results Source Data Save Load Polybase Enterprise Data Warehouse Azure SQL DW Azure Data Factory HDInsight Limitations  Expensive to keep it up & running  Slow to spin-up

| © Copyright 2015 Hitachi Consulting46 Spark ML on HDInsight Spark ML Pipelines Spark ML standardizes APIs for machine learning algorithms to make it easier to combine multiple task into a single pipeline, or workflow.  Transformers – used for data pre-processing. Input: DataFrame - Output:DataFrame  Estimators – ML algorithm used to build a predictive model. Input: DataFrame - Output: Model.  Parameters – Configurations for Transformers and Estimators  Pipeline – Chains Transformers and Estimators ML Pipeline Dataset (DataFrame) Transformer A (pre-processing) Estimator (ML Learning Algorithm) Model Evaluation Parameters Transformer Z (pre-processing) …

| © Copyright 2015 Hitachi Consulting47 Spark ML on HDInsight Spark ML Functionality Text Feature Extraction  TF-IDF (HashingTF and IDF)  Word2Vec  CountVectorizer  Tokenizer  StopWordsRemover  n-gram Feature Selection  VectorSlicer  RFormula  ChiSqSelector Dimensionality Reduction  PCA Features Vector Preparation  VectorAssembler  VectorIndexer  StringIndexer  IndexToString Transformers Feature Type Conversion  Binarizer  Discrete Cosine Transform (DCT)  OneHotEncoder  Bucketizer  QuantileDiscretizer Feature Scaling  Normalizer  StandardScaler  MinMaxScaler Feature Construction  SQLTransformer  ElementwiseProduct  PolynomialExpansion Estimators (supervised) Classification  Decision Trees – Ensembles  Naïve-Bayes  SVM Regression  Linear Regression  SVM Other (Unsupervised) Clustering Collaborative Filtering Frequent Pattern Mining

| © Copyright 2015 Hitachi Consulting49 Spark ML on HDInsight BigDL – Intel’s Distributed Deep Learning Library https://azure.microsoft.com/en-us/blog/use-bigdl-on-hdinsight-spark-for-distributed-deep-learning/

| © Copyright 2015 Hitachi Consulting50 Concluding Remarks Interactive Data Science Studio  Azure ML Extensibility  Spark on HDI  Azure ML  Microsoft R Server Built-in Features  Azure ML  Spark on HDI Rich Model Interpretability  SSAS Data Mining  Microsoft R Server Scalability (Big Data)  Microsoft R Server  Spark on HDI ML Pipelining  Spark on HDI  Azure Data Lake Analytics  SQL Server R Services  Data Mining SSAS Integration with Operational Apps  Azure ML  Azure Cognitive Services  Microsoft R Operationalization Pre-built Intelligence  Azure Cognitive Services  Azure Data Lake Analytics

| © Copyright 2015 Hitachi Consulting51 My Background Applying Computational Intelligence in Data Mining  Honorary Research Fellow, School of Computing , University of Kent.  Ph.D. Computer Science, University of Kent, Canterbury, UK.  28+ published journal and conference papers in the fields of AI and ML https://www.researchgate.net/profile/Khalid_Salama https://www.linkedin.com/in/khalid-salama-24403144/ https://github.com/khalid-m-salama/sqlbits-2017

Operational Machine Learning: Using Microsoft Technologies for Applied Data Science

More Related Content

What's hot

Similar to Operational Machine Learning: Using Microsoft Technologies for Applied Data Science

More from Khalid Salama

Recently uploaded

Operational Machine Learning: Using Microsoft Technologies for Applied Data Science

Editor's Notes