Inquiry About Multivariate Time Series and Covariate Support in Chronos

[Claire-bx]

Dear Chronos Team,

I hope this message finds you well. I have a couple of questions regarding the Chronos framework:

Does Chronos currently support multivariate time series prediction (i.e., multiple IDs simultaneously)?

I noticed that the latest ChronosX mentions the ability to use covariates as input, but I could not find guidance on how to use them. Does this mean that, at present, the models available for zero-shot or fine-tuning can only perform historical sequence → future sequence prediction and cannot incorporate covariate inputs?

Thank you very much for your time and assistance. I look forward to your guidance.

[Zambonilli]

autogluon has a higher level api that allows for covariates using regression and fine-tuning of chronos. Also, the predict method allows for multiple different "item_id" values to perform batch inference across multiple time series.

https://auto.gluon.ai/stable/tutorials/timeseries/forecasting-chronos.html#incorporating-the-covariates
https://auto.gluon.ai/stable/tutorials/timeseries/forecasting-chronos.html#fine-tuning

[abdulfatir]

@Claire-bx Chronos-2 supports univariate, multivariate and covariate-informed forecasting natively. Please check this notebook for details.

[Claire-bx]

does the latest Chronos-2 model support fine-tuning for the following tasks:
Univariate forecasting
Multivariate forecasting
Covariate-informed forecasting
In case I want to fine-tune the model, what kind of GPU resources would be recommended to run this efficiently?
Thank you very much for your help!

[abdulfatir]

@Claire-bx it's best to ask a question in the corresponding thread so it's easier for others to find the answer as well. @lostella has responded to this in #318.