AutoML for Time Series Forecasting

Although I'm not a fan of nowadays AutoML, I enjoy participating in AutoML competitions. I believe that such competitions are a great learning opportunity and could give strong pipelines for future projects. AutoML competitions are something in between Kaggle-like competitions and LeetCode tasks: your code should give high scores, and it should be steady and fast. Recently I took part in AutoSeries -- AutoML competition on time-series data, in which I managed to get the first place among 40 competitors (15 in the finals). This post is an overview of my solution.