Re: SNAPP gamma parameterization and prior specification

[Remco Bouckaert]

Hi Hunter,

The gamma distribution you displayed in BEAUti is paramterised as alpha=shape, beta=scale, with expectation shape*scale, not using alpha=shape, beta=rate with mean shape/rate. However, the SNAPP Prior (the one with alpha, beta, gamma, lambda entries in BEAUti) uses the alpha=shape, beta=rate. I realise that this must be terribly confusing, and an issues is logged to use only a single parameterisation (SNAPP issue #6) but we have not got around to fixing this yet.

Cheers,

Remco

On 3/12/2015, at 8:14 am, hunterA <richardhu...@gmail.com> wrote:

Anyone have insight into this? Perhaps there is something I am overlooking?

Thanks for the help!

-H

On Saturday, November 21, 2015 at 8:42:45 AM UTC-6, hunterA wrote:

Hi all,

I have a quick question concerning the prior specification of the gamma distribution on theta in SNAPP. In the original SNAPP paper, I see the following

Parameter, Distribution, Expectation

Theta, Gamma(shape = 1, scale = 200), 0.005

With the expectation, shape/scale = 0.005 

However, when I compare the gamma parameters for a different parameter (let's say lambda) I get widely different quantiles and expectation. 

See the attached photo "lambda_gamma_prior.png"

Additionally, when I plot the gamma distribution in R, I also get a widely different distribution.

See the attached photo "R_plotgamma.png"

I'm hoping someone could shed some light into this for me. Are these different parameterizations? and if so, is there a way to visualize the gamma prior distribution that I am setting up for theta in SNAPP correctly?

Thanks for the help and clarification!

-Hunter

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To post to this group, send email to beast...@googlegroups.com.
Visit this group at http://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.

[hunterA]

Hi Remco,

This makes sense now, thanks a lot for the response! 

For my dataset I have strong information that population sizes are 100,000 individuals with mutation rate = 10^-8 (theta = 4 * 100,000 * 10^-8 = 0.004)

So, given this information, I will set my prior settings for theta to alpha = 1 (shape) and beta = 250 (rate) for the gamma priors, with expectation = (1/250) = 0.004

For the lambda prior, I have strong prior information that my species have been diverging for a total of 20,000 generations, so I want a prior to reflect this expectation at

20,000 * 10^-8 = 0.0002 expected tree height.

I will try setting my lambda prior to 4,000 to reflect this expectation and see what I get.

Thanks a lot for the help!

Hunter

[Alexis Sullivan]

Hi Remco,

Is this fixed now?

best regards

Alex