4 years ago
Seven Reasons Budding Data Scientists Need a Machine Learning Course That’s Not Hands-On

From Coursera’s “Machine learning for Everyone”

My new course series on Coursera, Machine Learning for Everyone (free access), is for any learner who wishes to participate in the business deployment of machine learning, no matter whether you’ll play a role on the business side or the technical side. This end-to-end, three-course series is accessible to business-level learners and yet vital to techies as well. It covers both the state-of-the-art techniques and the business-side best practices.

Among machine learning courses, the three in this series are somewhat unusual. They’re not hands-on, and yet they do delve into the machine learning algorithms themselves, as well as other technical topics. So, if you’re a tech-oriented newcomer to machine learning who’s inclined to jump straight into a hands-on course, this article is for you. I wrote it to show you the value of taking courses like these before going hands-on – or even in addition to hands-on training after the fact.

On the other hand, if you’re not headed in that technical direction – if you’re not planning to ultimately do the hands-on number crunching yourself in your career trajectory – then this article will still help orient you and clarify the value of this course series.

And if you’re already a practitioner, past the learning phase, I invite you to consider this as a resource to recommend to executives, managers, and other colleagues in your circle you’d benefit by ramping up on machine learning from the inside out.

No hands-on. This course series does not include hands-on exercises in the operation of machine learning software, the core number-crunching activities. Rather than a hands-on training, it serves both business leaders and burgeoning data scientists alike with expansive coverage of the state-of-the-art techniques and the most prevalent pitfalls. There is no coding and no use of machine learning software.

But technical learners should take another look. Before jumping straight into the hands-on, as quants are inclined to do, consider one thing: This curriculum provides complementary know-how that all great techies also need to master. It contextualizes the core technology with a strong conceptual framework and covers topics that are generally omitted from even the most technical of courses, including uplift modeling (aka persuasion modeling) and some particularly treacherous pitfalls.

Seven vital topics skipped by hands-on courses

This course establishes foundational knowledge that, for data scientists, serves as a precursor to hands-on training. The fundamentals covered here are just as essential for hands-on practitioners as they are for business professionals, executives, and managers. Here are the topics we’ll cover that are more often than not skipped by hands-on ML courses:

1. Machine learning project leadership and deployment. Data scientists must learn to speak business language in order to jointly plan for and participate in a managed process that leads to ML deployment — which is the integration of a predictive model into existing operations. You’ve got to become fluent in how business objectives are conceived of and understand the organizational process within which the technology must be positioned.
2. Business-oriented metrics. How we report on the performance of predictive models in terms of organizational goals.
3. Foundational underpinnings. Unusually broad coverage of the conceptual principles behind ML algorithms.
4. Advanced methods. Certain key methods such as uplift modeling are very rarely covered, even in the most technical of courses, but are fundamental to achieving objectives in marketing and beyond.
5. Prevalent pitfalls. Key pitfalls such as p-hacking and “the accuracy fallacy” seldom receive the attention they deserve.
6. Data preparation. Hands-on courses typically begin with loading existing training data, skipping past its preparation. But data prep is central. It requires a lot of planning, business-side buy-in, and know-how.
7. Achieving equitable algorithms and other issues in ML ethics. In the final module of each course, we dive deeply into identifying risks to social justice and civil liberties that arise with a machine learning project, and presenting options to avert these risks. This includes machine bias, model transparency, explainable machine learning, and the right to explanation

Unlike most machine learning courses, this course series prepares you to avoid the most common management mistake that derails machine learning projects: jumping straight into the number crunching before establishing and planning for a path to operational deployment.

No matter how technically-oriented you are, jumping directly into the hands-on execution of machine learning can be a costly mistake, since doing so usually means skipping past these fundamentals. Ultimately, it’s all too easy as a techie to get stuck in the weeds, losing sight of the perspectives and priorities needed to make a real impact by ensuring that your models are actually applied — that is, they achieve their intended value-driven deployment.

It’s best to start with a holistic, business-oriented course on machine learning – no matter whether you’re more on the tech or the business side. After all, successfully deploying machine learning relies on savvy business leadership just as much as it relies on technical skill. Techies need to look beyond the number crunching itself and become deeply familiar with the business demands of machine learning.

Some learners want hands-on and only hands-on

Now, I certainly understand that many techies are dead set on jumping straight into the hands-on practice. I totally get it. Once upon a time, I was chomping at the bit just the same. If you’re certain about that, I say go for it! You can always return to this course series later to revisit the fundamentals you might otherwise miss.

But don’t be too dismissive of material like this designed to be friendly for non-technical learners. My hope would be that you understand that this level of curriculum does serve a critical need for its intended audience. After all, it’ll make your life as a predictive modeler a lot easier if the folks you work with on the business side are fluent with the concepts.

On the other hand, if you do enroll – and I hope you do – know that anyone destined to be a hands-on practitioner will need to continue on after this course with hands-on training. As with introductions to other technical areas, if you’re pursuing a career in the field, your first learnings set a strong foundation, deliver the prerequisite knowledge, and yet only whet your appetite for more. At the end of this course, I’ll guide those of you heading towards hands-on to select suitable options for further learning.

But there is one hands-on exercise

For one of the assessments of this course, you’ll perform a hands-on exercise, creating a predictive model by hand in Excel or Google Sheets and visualizing how it improves before your eyes.

However, other than that one hands-on exercise, there are no other assignments that have you executing on the number-crunching. That exercise operates within a spreadsheet so that most learners do not need to learn to operate any new software. This course series has no required exercises involving the use of machine learning software tools.

What this adds up to: your learning objectives

Let’s take a step back and look at the whole point of this course series.

After this course series, you will be able to:

• Participate in the application of machine learning, helping select between and evaluate technical approaches
• Interpret a predictive model for a manager or executive, explaining how it works and how well it predicts
• Circumvent the most common technical pitfalls of machine learning
• Screen a predictive model for bias against protected classes

This course will show you how machine learning works. It covers the foundational underpinnings, the way insights are gleaned from data, how we can trust these insights are reliable, and how well predictive models perform – calculating and reporting on the performance of a predictive model in business terms.

We’ll cover the standard algorithms, including decision trees, Naive Bayes, logistic regression, and neural networks – plus cutting-edge, advanced methods, such as deep learning, ensemble models, and uplift modeling (aka persuasion modeling). And we’ll prepare you to circumvent prevalent, treacherous pitfalls such as overfitting, p-hacking, and presuming causation from correlations.

I hope you enjoy and benefit from the course!

About the Author

Eric Siegel, Ph.D., is a leading consultant and former Columbia University professor who makes machine learning understandable and captivating. He is the founder of the long-running Predictive Analytics World and the Deep Learning World conference series, which have served more than 17,000 attendees since 2009, the instructor of the end-to-end, business-oriented Coursera specialization “Machine learning for Everyone”, a popular speaker who’s been commissioned for more than 100 keynote addresses, and executive editor of The Machine Learning Times. He authored the bestselling Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die, which has been used in courses at more than 35 universities, and he won teaching awards when he was a professor at Columbia University, where he sang educational songs to his students. Eric also publishes op-eds on analytics and social justice.