search

google-research / timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
https://research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting/
Apache License 2.0
3.81k stars 326 forks source link

Is it possible to extract the covariates weight? #106

Open lejarx opened 4 months ago

lejarx commented 4 months ago

I’m interested in the explainability of the forecast with covariates, or even univariate if possible.

Is there any method to extract the feature importance of the covariates.

The following notebook contains a walkthrough on covariates.

https://github.com/google-research/timesfm/blob/ec106a346c2471a5d21f1ea7ee0588fd0b41ccde/notebooks/covariates.ipynb

casaout commented 5 days ago

We are comparing TimesFM's ability to make short-term 1h forecasts against our existing pipeline. To that purpose, and to better understand the output, we'd also be interested in learning about how much each covariate (feature) contributed to the prediction. Is there already a way to do this? thanks!