This post is by Eric.

The next Stan meetup is coming up in February. It will be hosted by the New York Bayesian Data Analysis Meetup group and International Securities Exchange. The BDA group was formerly called Stan Users – NYC. We will still be focusing on Stan, but we would also like to open it up to a broader Bayesian community and hold more regular meetups.

P.S. What is Analogy Weighting you ask? I have no idea, but I am sure Jim Savage will tell us.

Erik do you know if they will somehow record the talk and put it on streaming?

+1

Even if not live, a youtube video would be very welcome.

Thanks for pestering me on this; I’ve been meaning to put together a short video demonstrating the method. I’m not sure whether Eric will be able to record the talk, but I’ll put together something anyhow.

Jim, thanks; that would be helpful.

We are not yet equipped to do live streaming, but this is something I will investigate later this year.

GIYF:

https://dl.dropboxusercontent.com/u/63100926/Thesis_final_Savage_James.pdf

This is Savage’s thesis work that should contain the results he is talking about

I should add about 20-30 years ago there was a fair amount of work in oceanography and meteorology on analogue based methods for predicting fields, a lot of it ad hoc (I would have to dig very hard to find these references). Also some related work in non-parametric methods to model time series. The name that comes to mind is S.J. Yakowitz, who was at Arizona, but I know there were others. Yakowitz’s approach was based on nearest-neighbor estimation. Interesting how ideas reappear over time and get improved on (btw – to be clear I am not commenting on Savage’s work, I haven’t had a chance to read his thesis, just that there is a history of approaching time series in this manner)

There was a spurt of activity around this in ecology a few years ago (Sugihara was one of the names). We’ve tried it and developed it a little on a few infectious disease prediction problems so I’m excited to hear about the method in another context. As far as I know the method goes by method of analogs/state space reconstruction/nearest neighbors but analogy weighting is a new one to me. Could we just pick a name?