a question about rjmcmc

Qing Zhao

unread,

Mar 11, 2025, 1:30:21 PM3/11/25

to nimble-users

Hi team,

I have the following for the indicator variable (z) and the corresponding slope (zbeta), which seems to be the standard way of using rjmcmc in Nimble:

[408567] RJ_indicator sampler: z_lambda0_br[1], mean: 0, scale: 0.20000000000000001, targetNode: zbeta_lambda0_br[1]

[408588] RJ_toggled sampler: zbeta_lambda0_br[11], samplerType: RW

My question is, is there a way to change the sampler of the slopes to something like RW_block_sampler (while maintaining rjmcmc)? Are there any benefits to do so? Please let me know. Thank you.

Best,

Qing

Perry de Valpine

unread,

Mar 11, 2025, 1:51:51 PM3/11/25

to Qing Zhao, nimble-users

Off the top of my head, I don't think so, because that would only make sense for a block of nodes toggled jointly, and it's not set up that way.

One could copy the source code for the samplers and start modifying for different or more general cases.

Perry

--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/nimble-users/d43216e3-a034-4a3a-95ed-04dfb91ac575n%40googlegroups.com.

Qing Zhao

unread,

Mar 12, 2025, 12:38:41 PM3/12/25

to Perry de Valpine, nimble-users

Dear Perry,

Thank you for letting me know.

Best,

Qing

Reply all

Reply to author

Forward