Slower sampling speed for LAN likelihood model

[zaizibai]

Dear HSSM developers,

I am trying to use the angel collapse boundary model with HSSM. However, i found MCMC sampling time is much slower than the standard DDM with analytical solution (I have 40 subjects, each finished 160 trials). Since LAN likelihood does not need to compute the infinite series in DDM analytical likelihood, it is reasonable to speculate LAN is faster than DDM analytical likelihood.

Best,

[frankmj]

Did you try this with GPU? Michael

…

On Mon, May 20, 2024 at 9:31 PM mqg ***@***.***> wrote: Dear HSSM developers, I am trying to use the angel collapse boundary model with HSSM. However, i found MCMC sampling time is much slower than the standard DDM with analytical solution (I have 40 subjects, each finished 160 trials). Since LAN likelihood does not need to compute the infinite series in DDM analytical likelihood, it is reasonable to speculate LAN is faster than DDM analytical likelihood. Best, — Reply to this email directly, view it on GitHub <#438>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAG7TFDBQ5ZHQS72Y4VTZQDZDKPXLAVCNFSM6AAAAABIAVA2QKVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZWG4YDENJRGA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[zaizibai]

Thanks for the response. No, i don't have a Nvidia GPU computer.

[digicosmos86]

Did you use the nuts_numpyro sampler?

[zaizibai]

I also tried the nuts_numpyro sampler, it did not speed too much.