search

steve-the-bayesian / BOOM

A C++ library for Bayesian modeling, mainly through Markov chain Monte Carlo, but with a few other methods supported. BOOM = "Bayesian Object Oriented Modeling". It is also the sound your computer makes when it crashes.
GNU Lesser General Public License v2.1
35 stars 14 forks source link

Add a continuous time local linear trend. #27

Open steve-the-bayesian opened 5 years ago

steve-the-bayesian commented 5 years ago

Are there some preferred ways of using bsts for "continuous time" time series i.e. when the intervals between samples are not equidistant? I guess I can time-box the samples and/or interpolate between the actual samples but if there are alternatives or you know of examples that would be great to know about.

Not presently, but they'd be easy (for me) to add. You'd just need a state model that was aware of timestamps. Timestamps are already a thing in bsts, but they're mainly used to identify time periods with multiple observations. How urgent is this request? I'm wrapping up the multiple time series thing, which is a major change that will benefit more people. But I'd be happy to add a continuous time local linear trend once that's done.

robertfeldt commented 5 years ago

Thanks, Steve, yes it is fairly urgent but not immediate. If a starting point for this is available in the April/May time frame I can test and give feedback etc.