create_jags_init function does not exist

[EmmaBenbow]

I dowloaded the gemtPlus package from GitHub a couple of weeks ago and it is working very well for most of the analyses that I am considering. However, there are one or two for which I run into convergence issues.

I would like to provide initial values for the betas for my fractional polynomials, to aid model convergence. I thought I should be able to do this using the create_jags_init function but, despite being listed as a function on GitHub, it seems that this does not exist?

(If there is a different way to set initial beta values, then I'd be grateful to know). Thanks

[MikeJSeo]

Hi,
I believe you cannot set initial values for non-variable values (i.e. beta) and can only set initial values to mu and d. Maybe you can set initial values for these to indirectly set initial values for the betas.
Anyways, I think here is a way of setting initial values (manually). Hope this works for you.

############################# Here's an example
library(dplyr)
library(gemtc)
library(gemtcPlus)
library(ggmcmc)

data("grouped_TTE")

inits <- list()
inits[[1]] <- list(d = matrix(c(NA, NA, NA,
0.44019988, -0.3689590, 0.02055352,
0.06173679, 0.1287214, -0.01493023,
0.56076502, -0.9448598, 0.12092799,
-0.01993132, -0.8312239, 0.13493946,
-0.44841103, 0.6927945, -0.07604329), ncol = 3, byrow = TRUE)
)
inits[[2]] <- list(d = matrix(c(NA, NA, NA,
0.44019988, -0.3689590, 0.02055352,
0.06173679, 0.1287214, -0.01493023,
0.56076502, -0.9448598, 0.12092799,
-0.01993132, -0.8312239, 0.13493946,
-0.44841103, 0.6927945, -0.07604329), ncol = 3, byrow = TRUE)
)

model_plan <- plan_fp(model.pars = list(exponents = c(0, 1), t.eval = "midpoint"),
bth.model = "FE", ref.std = "STUDY2", nma.ref.trt = "B",
n.chains = 2,
n.iter = 6000,
n.burnin = 1000,
n.thin = 1, descr = "Second order fractional polynomial model (fixed effect)"
)

# include inits here!!
model_plan$params$model_params$inits <- inits
#model_plan$params$model_params$parameters <- c("d", "mu", "Beta")

model_input <- nma_pre_proc(grouped_TTE, model_plan)

model <- nma_fit(model_input = model_input)

[EmmaBenbow]

Hi. Thanks so much for looking into this and your reply - I will give this a go!
Best wishes