Hello! I have a question about detecting multiple anomalies or changes in time series trends. Is this something that's possible with TimeGPT or our other packages? Thanks so much.

Leon-Friedrich Thomas

Email: leonfriedirchthomas@gmail.com

I want to use a time series of multiple variables, which are assessed at the same time. And have questions regarding that: Is it possible with the api to not only detect single outliers but also outlier periods where the latest period e.g. year or 7 months doesn't fit the historic data? Is there any good solution for this use case available ? The aim of this would be to detect not only single outlier values but a change in the time series trend.

I think I have a draft answer to this question: 1. Multi-Series Anomaly Detection: We're actively working on this feature, and we hope to release it with TimeGPT 1.5. 2. Outlier Period: TimeGPT is capable of detecting the beginning of a trend shift. However, after the shift period stabilizes, TimeGPT will recognize the new trend as normal.

@Tracy Teal, as @Yibei mentioned this is a feature we are actively developing. It would be great if we further understand the use case of Leon. Would you mind asking him what he is working on? Maybe also company, etc.

Yes, sure I'll follow up.

and we can send them the wrapper as well for testing if they are already in the trial

How do we send people the wrapper?

Someone else wants it? Yibei is still working on it, an example, and a loom video explaining it

oh, I see, we see if they want to join the trial? And then we would send him the wrapper?