-
Notifications
You must be signed in to change notification settings - Fork 21
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Gaussian shells convergence #34
Comments
OK, I made some progress. I realised that I need to draw from the entire prior, because the prior to posterior shrinkage is constrained by how particles go up in levels. I also needed to modify the perturb function to cover more orders of magnitudes, with:
|
I also had to increase to using 100 levels. |
Are you happy with it now? Your proposal covers a very wide range of
scales. These days I use a log cauchy for that. Also you might try letting
it choose the number of levels itself but if there's a phase change that
might work as well.
…On Tue, 30 Jun 2020, 8:46 AM Johannes Buchner, ***@***.***> wrote:
I also had to increase to using 100 levels.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#34 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAMBKOTTFDTFNB6E4TAHU3TRZD4TJANCNFSM4OLMCLAA>
.
|
Hi,
I wanted to show a small demo of Diffusive Nested Sampling in comparison to other techniques. I set up two gaussian shells which work as a spike-slab model in 2d. The posterior looks like a diamond ring (the second peak/shell is the knot on the left):
I thought this problem is nice because it tries to mimic some problems of high-d -- MCMC auto-correlation is rather poor, it has heavy tails, etc. I set up DNest4 code, and it runs, but it converges extremely slowly towards the true lnZ (-22.93).
Output:
My questions are:
Code:
The text was updated successfully, but these errors were encountered: