Executive Guide to Predictive Modeling Strategy at Scale

Course details

Building world-class predictive analytics solutions requires recognizing that the challenges of scale and sample size fluctuate greatly at different stages of a project. How do you know how much data to use? What is too little, what is too much? How does your infrastructure need to scale with the volume and demands of the project? This course walks step by step through the strategic and tactical aspects of determining how much data is needed to build an effective predictive modeling solution based on machine learning and what volumes of data are so large that they will create challenges. Instructor Keith McCormick reviews each stage—data selection, data preparation, modeling, scoring, and deployment—with scalability in mind, providing IT professionals, data scientists, and leadership with new insights, perspectives, and collaboration tools.

Note: This course is software agnostic. The emphasis is on strategy and planning. Examples, calculations, and software results shown are for training purposes only.

Learning objectives
Evaluating the proper amount of data
Assessing data quality and quantity
Seasonality and time alignment
Data preparation challenges
Data modeling challenges
Scoring machine-learning models
Deploying models and adjusting data prep and scoring
Monitoring and maintenance

Concepts

0. Introduction

01 - Scaling machine learning initiatives
02 - Defining terms

1. The Phases of a Machine Learning Project

03 - Data and supervised machine learning
04 - The nine big data bottlenecks
05 - The stages of predictive analytics data
06 - Why you might have too little data

2. Designing a Machine Learning Dataset

07 - How much data do I need
08 - Balancing
09 - Who truly has big data
10 - Assessing data
11 - Selecting - Data that should be left out
12 - Seasonality and time alignment

3. Data Prep Challenges

13 - Data and the data scientist
14 - Aggregate and restructure
15 - Dummy coding
16 - Feature engineering

4. Modeling Challenges

17 - Understanding the modeling process
18 - Slow algorithms - Brute force
19 - Slow algorithms - More calculations
20 - Slow algorithms - More models
21 - How to sample properly
22 - Modeling with missing data

5. Scoring

23 - Scoring traditional ML models
24 - Scoring a black box model
25 - Scoring an ensemble

6. Deployment

26 - Batch vs. real-time scoring
27 - Data prep and scoring
28 - Combining batch and real-time scoring

7. Monitoring and Maintenance

29 - What is model monitoring
30 - How often should you rebuild

Conclusion

31 - Next steps