Michael, this piece is awesome. Simple yet effective as an intro to Time Series piece. I was thinking about a critique i read on a 2019 paper about ML models for time series, and then discuss issues with legacy models like ARIMA and how they are limited to historical data and don'\t deal well with the Cold Start Problem, and how models like the DeepAr from Amazon are actually a better approach to Forecasting. Any thoughts on how to know when to use a time series approach to a time series problem and when to use a regression approach, maybe with some deep learning flavouring? Cheers and thanks for the food for thought!