IEPDS: Immersion Event on Practical Data Science with Cortana Analytics: Azure Machine Learning, SQL Data Mining and R
(Retired – no longer offered as a public class)
Overview
This 5-day class is the first of the two data science courses taught by Rafal Lukawiecki. Some of the topics will be introduced at level 200, without requiring you to have data science prerequisites, on the first day. However, as the class progresses, the level of the training will quickly increase to 300 and 400.
You will learn machine learning, data mining, some statistics, data preparation, and how to interpret the results. You will see how to formulate business questions in terms of data science hypotheses and experiments, and how to prepare inputs to answer those questions. We will cover common issues and mistakes, how to resolve them, like overtraining, and how to cope with rare events, such as fraud. At the end of this course you will be able to plan and run data science projects.
As a practicing data miner, Rafal will also share his decade of hands-on experience while teaching you about Azure Machine Learning (Azure ML) which is the foundation of Cortana Analytics Suite, and its highly-visual, on-premise companion, the SQL Server Analysis Services Data Mining engine, supplemented with the free open source and Cortana’s Revolution Analytics R software. We will use some Excel, however, most of our time will be spent in ML Studio, some in R, RStudio, SSDT, SSMS, and the Azure Portal.
Prerequisites: No formal prerequisites. Basic knowledge of SQL Server Data Tools, Excel and any analytical experience helps. Best of all: prepare questions that you would like to answer using predictive analytics and machine learning.
Format: 60% lectures, 20% demos, plus 20% time allocated to help you follow the demos and tasks on your own equipment, if you bring a laptop. You will be challenged to find answers to 4 problems during the course, and you will have a chance to build your own models in SSAS, Azure ML and R. Doing that will help you learn, however, it is not a requirement: you are welcome to observe the demos and ask questions, too. If you bring your own data, you are welcome to analyse it, too. You will get a list of free or evaluation-edition software to preinstall before attending. You will need your own Azure account: free one is OK, but the paid one is better—and it can be inexpensive, or even free during a trial. You can copy course experiments and data into your ML workspace for learning and future reference.
Target audience: Analysts, power users, predictive and BI developers, database and other professionals who wish to embrace machine learning, budding data scientists, consultants.
Instructor: Rafal Lukawiecki (You can watch a 75-minute presentation on this material by Rafal at the 2015 Ignite conference here.)
Need Help Justifying Training? Here’s a letter to your boss explaining why SQLskills training is worthwhile and a list of community blog posts about our classes.
Quotes From Past Attendees
Listed below are some verbatim quotes from recent attendees of this class:
- “Rarely do we meet domain expertise, in any discipline, that is as complete and deep as Rafal’s in Data Science/Data Mining/R, SSAS, AML. Add to the observed level of patience and ability to articulate and concern for the class’ absorption of the material and ‘lucky/fortunate’ is the best way to describe anyone who attends one of Rafal’s training events.”
- “Would come back again – just loved it!”
- “Rafal was just great!”
- “‘Boots on ground’ – real-world applicable.”
- “Ad hoc deep questions answered with real-life examples or academic explanations softened with real-world perspective.”
Curriculum
Please note: we reserve the right to amend the order and the day allocation of the modules to best suit the dynamic character of the class and to answer questions as they arise. Some subjects (marked with an asterisk *) are optional, and will only be covered if time allows.
Module 1: Overview of Practical Data Science for Business
We begin the course with a quick, high-level introduction of all of the key concepts, terminology, components, and tools. Topics covered include:
- Introduction to data science and its components
- Machine learning vs data mining
- Statistics
- Big data
- Data wrangling
- Team, process, and tools
Module 2: Tools and Getting Started
Configuring Cortana Analytics key component, cloud-based Azure ML is effortless. You need to pay a little bit more attention to on-premise R and SQL server environment, to make sure that you can easily access your modeling data. Topics covered include:
- Getting started with and using Cortana Analytics: Azure ML, SSAS DM, and R
- Structures, models, data flows
- Configuration concerns and pricing
- Azure requirements and dependencies
- Other components of Cortana Analytics Suite
- Using Rattle with R and RStudio
- Getting a feel for the data: interpreting notched boxplots in R
Module 3: Data
Data science requires you to prepare your data into a rather unique, flat, and completely denormalised format. While inputs are always necessary, and you may need to engineer hundreds of them, we do not need predictive outputs in all cases. Topics covered include:
- Inputs and outputs, features and labels
- Data formats, discretization vs. continuous
- Cases, observations, signatures
- Feature engineering
- Azure ML data preparation and manipulation modules
- Preparing unstructured text for text analysis
- Feature hashing
- Moving data around and its storage
Module 4: Process
The analytical process consists of problem formulation, data preparation, modelling, validation, and deployment—all in an iterative fashion. You will learn about the CRISP-DM industry-standard approach, as well as the application of the scientific method of reasoning to experimentation, when solving real-world business problems. Topics covered include:
- Stating business question in data science term
- CRISP-DM
- Scientific method of reasoning
- Hypothesis testing and experiments
- Iterative hypothesis refinement
Module 5: Algorithm Overview
There are hundreds of machine learning algorithms, yet they belong to just a dozen of groups, of which 4-5 are in very common use. We will introduce those algorithm classes, and we will discuss some of the most often used examples in each class, while explaining which technology tools (Azure ML, SQL, or R) provide their most convenient implementation. Topics covered include:
- What does data mining do?
- Algorithm classes in Azure ML, R, and SSAS
- Supervised vs Unsupervised learning
- Classifiers
- Clustering
- Regression
- Similarity Matching
- Recommenders
Module 6: Segmentation
Segmentation is the main application of unsupervised learning using clustering algorithms. While the action of the algorithm is usually quick and easy to configure, interpreting the results can take a lot of time and intuition. We will spend plenty of time practicing segmentation, interpreting the results and subsequently parameterising the algorithm to provide us with additional insight, and to help you apply it back to your own data. You will even learn how to apply this technique for anomaly (outlier) detection and text analytics! Topics covered include:
- Introduction to segmentation
- Clustering algorithms (k-means, EM, and others)
- Interpreting clusters
- Cluster characteristics
- Discrimination
- Tornado charts
- Using clustering for text analysis
- Anomaly detection with clustering, PCA and SVMs
Module 7: Classification
Without doubt, classifiers are the most important, and the most often used category of machine learning algorithms, and the foundation of algorithmic data science. We will focus on several variants of the most important classifier algorithm—decision tree—while progressively interpreting the results, and improving its performance. After introducing neural networks and logistic regression we will also compare the performance of all of these classifiers on our test dataset. Topics covered include:
- Introduction to classifiers
- Two-class (binary) vs multi-class
- Decision trees, forests, and boosting
- Decision jungles *
- Neural networks and logistic regression
- Overfitting (overtraining) concerns
- Using classifiers for text analysis
- Associative decision trees *
Module 8: Basic Statistics
Basic concepts of statistics, notably: means, medians, modes, and variance or standard deviation, are essential to validating data and model quality. Probability, and the concept of p-values help you decide which of your inputs (features) are more important than others. R makes all of these powerful ideas accessible and visual, while Azure ML enables you to deploy them easily into production. Topics covered include:
- Basic concepts of statistics: population vs sample, measure types, means and deviations, distributions, confidence intervals, p-values
- Correlation
- Descriptive statistics with R
- Basic concepts of probability
- Finding important features using p-values, linear regression, and ANOVA *
Module 9: Model Validation
The most important aspect of any data science project is the iterative validation and improvement of the models. Without validation, your models cannot be used. There are several tests of model validity, and we will focus on accuracy and reliability, showing you different ways to measure it. Topics covered include:
- Testing accuracy
- Lift charts
- Testing reliability
- Testing usefulness
Module 10: Classifier Precision
Validation of classifiers is likely to be your main occupation as a data scientist, because classifiers are used so often, and because their precision is not always easy to balance with business requirements, such as restricted resources or required business performance. We will introduce the fundamentals of finding the balance between the acceptable number of false positives and false negatives by using classification (confusion) matrices, and plotting the options using ROC (Receiver Operating Characteristic) charts. Topics covered include:
- Testing classifiers
- False positives vs. false negatives
- Classification (confusion) matrix
- Precision
- Recall
- Balancing precision with recall vs. business goals and constraints
- Charting precision-recall (sensitivity-specificity)
- ROC curves
- Other measures of accuracy
- Cross-validation
- Optimizing binary classifier thresholds for a known business goal of a prediction quality
- Refining models to improve accuracy and reliability
- Using parameter sweeps to fine-tune algorithm performance
- Class imbalance problem (fraud analytics and rare event prediction) *
Module 11: Regressions
Considered by some as the numerical equivalent of classifiers, regression is a large subject of its own. We will introduce its simple but a very popular form, linear regression, and the more precise, but also prone-to-overfitting, decision tree variant. Topics covered include:
- Introduction to simple regressions
- Linear regression (classic)
- Regression decision trees and other ensemble regression algorithms
- Relationship to ANOVA *
- Measuring linear regression quality (R-squared, predictor p-values, RMSE, MAE, RAE, RSE, and additional testing using R) *
Module 12: Similarity Matching and Recommenders
From basic concepts of similarity matching, through model-based associative analysis, collaborative filtering, to hybrid systems, like the Matchbox algorithm, there are several techniques for building recommenders. You will get a good overview of this subject, as well as an understanding of how to use these techniques for advanced data exploration, such as Market Basket Analysis. Topics covered include:
- Introduction to recommender concepts
- Model-based, similarity-based, and hybrid recommenders
- Association rules
- Understanding itemsets and rules
- Rule importance vs. rule probability
- Data structures for association rules
- Market Basket Analysis
- Collaborative filtering
- Matchbox recommenders
- Validating recommenders
Module 13: Other Algorithms (Brief Overview)
As the course is coming to its end, we will briefly overview some of the remaining and interesting algorithms and techniques, such as text analytics, without going into too much detail, but letting you have an understanding of the existing approaches. Topics covered include:
- Text analysis
- Sequence clustering and Markov chains
- SVM (Support Vector Machines)
- Time series
- Image analytics
Module 14: Production and Model Maintenance
If you plan on using your models for prediction, rather than just for the exploration of data, you need to deploy your models to production and maintain them on an on-going basis. You will learn about the easiest way to do so using Azure ML web services and its REST synchronous and asynchronous APIs, as well as how to deploy and invoke SSAS models by using DMX queries. Topics covered include:
- Deploying models to production
- SSAS models and DMX queries
- Azure ML web services: preparation and publishing
- Cortana Solutions Gallery
- REST APIs: request/response vs. batch
- On-going maintenance and model updates
Questions?
If you have any questions not answered by our F.A.Q., please contact us.