Our quest for robust time series forecasting at scale
Data Science Time Series Machine Learning AutoML ZettelkastenAn older link from April 2017 that I believe became AutoGluon. AutoGluon is fantastic for time series and a host of other AutoML use cases.
So, what models do we include in our ensemble? Pretty much any reasonable model we can get our hands on! Specific models include variants on many well-known approaches, such as the Bass Diffusion Model, the Theta Model, Logistic models, bsts, STL, Holt-Winters and other Exponential Smoothing models, Seasonal and other other ARIMA-based models, Year-over-Year growth models, custom models, and more. Indeed, model diversity is a specific objective in creating our ensemble as it is essential to the success of model averaging. Our aspiration is that the models will produce something akin to a representative and not overly repetitive covering of the space of reasonable models. Further, by using well-known, well-vetted models, we attempt to create not merely a “wisdom of crowds” but a “wisdom of crowds of experts” scenario, in the spirit of Mannes [6].