Welcome to the third week of the "How to Win a Data Science Competition" course! Here is a short summary of what you will learn.

• Dmitry Ulyanov will start this week with analysis of evaluation m etrics . We will recap the most prominent ones for tasks of classification and regression, examine them for the best constant predictions to create simple baselines in competitions, and then see, how we can efficiently optimize a specific metric in a particular competition.

• Dmitry Altukhov will discuss very powerful technique for feature generation - mean encodings . We will start with categorical features, analyze how mean-encoded features lead to overfitting and overview how can we balance overfitting with regularization. Then, we will discuss several extensions to this approach, including applying mean-encodings to time series, numeric features and combinations of features. In the programming assignment you will need to implement several regularization techniques for mean encodings.

To keep up the work on the final project , apply ideas we will discuss and improve your current solution by checking the metric you optimize and by adding mean-encoded features.

Now let's go ahead and get started!