search

university-of-newcastle-research / pmwg

Repository for code for the Samplers Team at UoN
3 stars 3 forks source link

Added epsilon per person #64

Closed niekstevenson closed 1 year ago

niekstevenson commented 1 year ago

This epsilon per person is flexibly tuned by a new argument p_accept. Whatever you set p_accept to, that value the acceptance ratio for each person will approximate the acceptance rate of new particles.

gjcooper commented 1 year ago

Hey Niek,

I noticed when you call the update_epsilon function, you pass in the subj_epsilon squared, as in:

update_epsilon(subj_epsilon^2, acc, p_accept, j, pmwgs$n_pars, alphaStar)

But then inside the update_epsilon function you take the square root:

Theta=log(sqrt(epsilon2))

I was wondering whether there was any particular reason why you do this?

niekstevenson commented 1 year ago

Hey Gavin,

Good catch. I'm pretty sure that's not necessary, because it would never return negatives.

Cheers, Niek

Op di 14 feb. 2023 om 02:04 schreef Gavin Cooper @.***>:

Hey Niek,

I noticed when you call the update_epsilon function, you pass in the subj_epsilon squared, as in:

update_epsilon(subj_epsilon^2, acc, p_accept, j, pmwgs$n_pars, alphaStar)

But then inside the update_epsilon function you take the square root:

Theta=log(sqrt(epsilon2))

I was wondering whether there was any particular reason why you do this?

— Reply to this email directly, view it on GitHub https://github.com/university-of-newcastle-research/pmwg/pull/64#issuecomment-1428943222, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALR27B66CE2SBCNXUAW7PNTWXLKZ5ANCNFSM6AAAAAAUF4VZ6I . You are receiving this because you authored the thread.Message ID: @.***>