Add exogenous variables or finetuning model

#2
by lorenzolaudato - opened

Hello,

is it possible to add future/past known covariates (exogenous variables) for the prediction or finetuning model on own data ?

Thanks

MLDI Lab @ ShanghaiTech University org

Hi @lorenzolaudato ,

Thank you for your question.

Similar to most time series foundation models, Kairos currently only supports channel-independent forecasting (like PatchTST). This means that each variable in a multivariate time series is forecasted independently.

Therefore, while you can finetune the model on your own data, the model cannot capture the relationships between variables and will only predict each one independently. We are actively considering the modeling of inter-variable relationships in our future work.

Best regards.

Thank you for your answer. Is it possibile to have a little tutorial on how finetune model on own data ? Thanks, Lorenzo

Sign up or log in to comment