Coursera: How to Win a Data Science Competition Review

How to Win a Data Science Competition: Learn from Top Kagglers offered by Russia's National Research University Higher School of Economics as a part of its advanced machine learning specialization is a novel data science course that teaches tips and tricks to rank highly in machine learning competitions. The five-week course covers a wide range of topics related to improving predictions of machine learning models, such as feature preprocessing, feature engineering, validation, optimization, parameter tuning and ensembling. The course page states that you should know the basics of Python and its data science packages like pandas and scikit learn as well as the basics of machine learning before taking the course. I'd also recommend having at least some familiarity with data science competitions, such as by participating in some of the educational completions on Kaggle, to get the most out of the course. A lot of the course content, including quizzes and programming assignments, requires a subscription to access so this review will be based primary on the lecture content. It does offer a final project in the form of a Kaggle competition that all users can access.

How to Win a Data Science Competition is taught by an an ensemble cast of 5 lecturers who have placed highly in various competitions on Kaggle, the leading website for data science competitions. All the lecturers are top 100-level Kaggle masters, so they have a wealth of knowledge and practical hands-on experience with the subject. Each week of the first four weeks is packed with 2-3 hours worth of lecture content, while the final week discusses approaches in highly scoring submissions to several past Kaggle competitions. The lecture video quality is good and the instruction is adequate, but be aware that the course is taught by a bunch of data science nerds who don't speak English as a first language. This may make it difficult for some learners to follow along at times considering the subject matter itself isn't always easy to understand. Despite its shortcomings, the course offers a lot of value to anyone interested in data science competitions as it covers many topics and tricks you won't find in other MOOCs, like exploiting data leakages and using model stacking to improve predictions. It also spends a lot of time discussing advanced feature engineering, which is often a key to winning competitions.

How to Win a Data Science Competition is a unique course that offers a treasure trove of techniques to help students place higher in data science competitions and sheds light on some of the methods top Kagglers use to get an edge on the competition. Although certain topics covered in this course are not directly applicable to real world projects, seeing the methods and ingenuity that go into winning competition solutions is enlightening.

I give How to Win a Data Science Competition 4.5 out of 5 stars: Great.