N-Beats
Data Skeptic
00:00
The State of the Art in Machine Learning for Time Series Data
The state of the art was the model developed by smill, who actually participated in the competition. And his model was a heavy ensamble, i think it's like 90 models in ansample. It was based on a combination of the recurrent neural network and the classical approach. The classical approach was used to remove trend and somehow normalize the data. A residual stack of arenan models was used to model the residuals from that.
Transcript
Play full episode
