Time Series¶
methods¶
https://news.ycombinator.com/item?id=37877443
On extremely high dimensional data:
MLPthat just uses lagged values as featuresOn mid-dimensional data:
LightGBM/Xgboostis by far the best and generally performs at or better than any deep learning modelOn low-dimensional data:
(V)ARIMA/ETS/Factormodels are still king
model history¶
https://medium.com/@ycwong.joe/a-brief-history-of-time-series-models-38455c2cd78d
summary¶
https://chartexpo.com/blog/time-series-forecasting
How to feed forecast into next step's input feature¶
https://www.kaggle.com/code/ryanholbrook/forecasting-with-machine-learning
multioutput model
each model for one step
one model, update feature based on prevous forecast
each model for one step, update feature based on prevous forecast