Welcome to the blog
All additional materials for the articles can be found on the Github.
Car rental problem
I'd like to present to you the problem I stumbled upon in the R. Sutton and A. Barto book about reinforcement learning and really liked. You manage two locations for car rental and try to maximize your profit by moving available cars around locations. It can be seen as a very close model to the real-world situation. It is also very convenient to show how to approach and solve such problems in terms of the RL.
What is the weather and can it be predicted?
This is an overview post that defines the weather from the physical point of view, gives a theoretical framework for weather modelling and explains why we can't build long-term weather forecasts. It also introduces classical and ML approaches for prediction and shows how they can be blended together.
This post can be a good introduction for anyone curious about how their weather forecast is made. The reader is expected to have a knowledge of university-level calculus and general physics.
On different interpretations of the class weighting and notion of unbalanced classes
Class weighting is often referred to as a simple and powerful technique to use when solving classification problems with unbalanced classes. But how can it be interpreted in a probabilistic sense? We are going to answer it using binary classification problem with the logistic regression model as an example. However, the more general question of the article is actually why unbalanced classes is the issue in the first place? Is it the issue by itself or is it a conseqence of some underlying property of the data? We will figure this out by the course of the narrative.