Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Hello everyone,
I'm currently using the `statsforecast` library to perform distributed training on Spark for approximately 100k time series. For each series, I'm fitting the MFLES, ARIMA, and SeasonalNaive models. I was wondering if there are any plans to include ensemble techniques in the near future, such as averaging multiple models together for better, more accurate predictions, or even weighted ensembles based on error metrics.

Hi <@U066NNB5V71> thank you for using statsforecast, In the short-term, we don't have plans to include ensemble techniques.

I'll recommend you try to leverage Spark and statsforecast. For example, getting a simple average is relatively straightforward, since you can just multiply the output rows.

regarding the weighted ensembles, those can be useful,  but usually it's hard to beat the simple average as mentioned <https://otexts.com/fpp3/combinations.html|here>.