Timeline for Weighting observations and measurement uncertainty in bayes
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 13, 2015 at 4:41 | vote | accept | StevenMurray | ||
| Jul 13, 2015 at 4:41 | comment | added | StevenMurray | I have tagged your answer as "correct", because it was a correct answer to my question, though ultimately not helpful (but that was my fault). I have a more thorough question on this now: stats.stackexchange.com/questions/161134/… | |
| Jul 6, 2015 at 6:48 | comment | added | StevenMurray | Also, I figured out why this example wasn't working, and I have fixed it. The main thing wrong was that I was using $\alpha+s$ rather than $\alpha-s$ -- a carry-over from the actual use-case. I've posted the code that works in my question as an edit. The real question then is: why doesn't this work in the truncated GGD case? I also had a look at the discussion on the weighting, and agree that no exact generative model can be made: but pragmatically it seems to work okay. | |
| Jul 6, 2015 at 6:42 | comment | added | StevenMurray | I see that that would work, but unfortunately for my actual use-case, it won't (in my actual use-case I'm not using a Pareto distribution, but rather a truncated generalised gamma distribution). Furthermore, the power-law slope of the distribution is $< -1$, which means that it only converges because of the truncation. Unfortunately, the Boost library only supports the incomplete gamma function with $z>0$, where $z=(\alpha+1)/\beta$, so I can't use it as is. This "weighting" has as its primary purpose to modify the slope so I can actually use Stan. | |
| Jul 4, 2015 at 18:16 | history | edited | Ben Goodrich | CC BY-SA 3.0 |
reparameterized to use log_minimum rather than minimum
|
| Jul 4, 2015 at 17:36 | history | answered | Ben Goodrich | CC BY-SA 3.0 |