Welcome to this short, one-week course, which covers the most neglected yet critical skills in machine learning, four vital techniques that are almost never taught -- most courses and books omit them entirely. Okay well one's more of a pitfall that doesn't actually require much skill to avoid once you're aware of it, but whatever -- four absolutely crucial special topics. 1) Uplift modeling (also known as persuasion modeling) which predicts not the outcome or behavior but your influence on that behavior. A total paradigm shift for machine learning that can deliver a tremendous improvement for marketing, personalized medicine, and political campaigns. Standard methods don't attempt to model influence and therefore aren't designed to optimally predict the thing you care about, which is, will this action have the desired effect? Like a marketing treatment -- will it cause a purchase? Amazingly, most data scientists don't know about uplift modeling. 2) The accuracy fallacy - An extremely common pitfall where you effectively lie -- to others if not also to yourself -- about how well a model predicts. 3) P-hacking - Another common pitfall where you draw a false conclusion from data. Bad science. Avoiding this one takes some real technical know-how. 4) The paradox of ensemble models - Why do ensembles generalize well rather than overfitting, given that they just pile on complexity? Now, it's crazy that these four topics are so rarely covered. They're essential because they address questions that are super fundamental to machine learning: Uplift modeling: When you're modeling, are you even predicting the right thing? The accuracy fallacy: When evaluating how well a model works, are you even reporting on the right thing? P-hacking: Are your simplest discoveries from data even real? The paradox of ensemble models, which are the most elegant way to advance modeling beyond simple methods like decision trees and logistic regression without taking on the often-unnecessary overhead of deep learning. So, do you understand how they work, even though they seem to defy Occam's Razor, the preference for simplicity? For most data scientists and for most machine learning projects, the answer to each of these questions is quite often no. Averting the major pitfalls is the lifeblood of machine learning because to avoid them is to get machine learning to actually work. And many projects won't succeed without uplift or ensemble modeling. Now, to assemble this one-week course, I sourced the videos from my much longer machine learning specialization, which is a comprehensive series of three courses. It covers both the technical and business sides of machine learning, but it takes 12 weeks. In contrast, this course pares that down to the most vital yet rarely-covered single week of material -- on only the technical rather than business side -- although I should note that you're free to spread it out over multiple weeks or even binge it all in one day. This comes to 11 videos, not including this video and a closing video, plus practice quizzes and optional further reading throughout and a graded quiz at the end. By the way, the videos are not in this format shot outside; they're in this format. This course does assume you're already familiar with machine learning as a field and at least the standard modeling methods such as decision trees. Now, unlike many technical machine learning courses, this one does not include hands-on training. There's no coding and no use of machine learning software. Instead, this short conceptual course lays the groundwork before you take on the hands-on practice. Data scientists often want to get right to the hands-on, but, when it comes to these state-of-the-art techniques and prevalent pitfalls, there's a foundation of conceptual knowledge to build first and you'll be glad you did. By the way, if you're focused on deep learning, this course does apply -- because most machine learning fundamentals apply just as much when employing deep learning as with other machine learning methods. The accuracy fallacy actually comes up more with deep learning projects than with other machine learning methods. And once you see the importance of uplift modeling, you'll understand that the power of deep learning should be applied there as well, although as of this recording, to my knowledge that's not been tried yet. It's definitely coming. Ensemble models are often an important alternative to deep learning. The "deep learning" craze tends to undervalue traditional machine learning methods, but since deep learning is overkill for many projects, being well-versed in the leading method to amp up capabilities without involving the tremendous overhead of deep learning -- ensembles -- is an absolute must-have for every data scientist's toolkit. Now, it's been a great privilege to partner with the renowned analytics company SAS in order to bring you this course. SAS is a long-standing industry leader who's been at the forefront of delivering machine learning capabilities for more decades than any other vendor I know. On the other hand, this curriculum is vendor-neutral. The concepts and the learning objectives apply, regardless of which of the many machine learning software tools you might end up working with. I developed this material independent of my partnership with SAS. SAS will augment this course with optional demo videos that helpfully illustrate some of the concepts in action. But you don't need to watch those videos in order to prepare for the graded assessment, so you can pick and choose at will along the way. Okay, we're ready to begin on these rare skills. These are special topics, but they shouldn't be -- they really should be considered fundamentals by the industry. All data scientists should master them. Someday they'll be standard -- in the meantime, this course will leapfrog you ahead and give you an advantage. I'm sure that, throughout your career, you'll find these skills indispensable. But first, a brain teaser: What often happens to you that cannot be perceived and that you can't even be sure has happened afterward -- but that can be predicted in advance? Good luck and if you get stuck, you'll learn the answer during the videos on uplift modeling.
