zhaokg / Rbeast

Bayesian Change-Point Detection and Time Series Decomposition
208 stars 36 forks source link

Dos it support online mode? #5

Open bigearmouse opened 1 year ago

bigearmouse commented 1 year ago

This is great work. I know the origional paper from Adam and MacKay talking about online changepoint. Does it support online calculation? thanks.

zhaokg commented 1 year ago

Thanks a lot for asking. For your quick reference, the algorithm implemented here has nothing to do with the Adam and MacKay paper. Sorry about the confusion. One major difference is that the beast algorithm seeks to decompose time series into trend and seasonality. Also, the algorithmic heuristics are different. More importantly, beast is an offline algorithm. Recently, some researchers used BEAST for online changepoint detection, and it worked. Let me know if there are any specific questions.

bigearmouse commented 1 year ago

thanks for your response. previously I had trouble openning your paper. But I have access now. Will read the paper first. Thanks for your help.

gwerbin commented 1 year ago

some researchers used BEAST for online changepoint detection

Hi @zhaokg , this sounds very useful. Do you know if this has appeared in any published work yet? I would love to see how they used it.

zhaokg commented 1 year ago

some researchers used BEAST for online changepoint detection

Hi @zhaokg , this sounds very useful. Do you know if this has appeared in any published work yet? I would love to see how they used it.

Hi @gwerbin , this is a paper just published today: https://www.sciencedirect.com/science/article/pii/S0924271623000424?via%3Dihub, which used the BEAST algorithm for continuous online detection. But I have to say that I did a sloppy job in implementing BEAST and it can be improved or revised to better accommodate online changepoint detection. If you see some good values of BEAST for your application, you can please share an example dataset for me to test-run and I will test the possibility of revising the algorithm for more efficient online detection.