Inquiry Regarding Model Errors in NIMBLE
B Motwaly
Dear NIMBLE Group,
I hope this message finds you well. I am currently working on a population dynamics model using NIMBLE and have encountered several warnings that I would like your guidance on:
Backwards Indexing Warnings:
- [Warning] Detected backwards indexing in '10:9'. This is likely unintended and will likely not produce valid model code.
- [Warning] Detected backwards indexing in '12:11'. This is likely unintended and will likely not produce valid model code.
- [Warning] Detected backwards indexing in '14:13'. This is likely unintended and will likely not produce valid model code.
Non-Constant Indexes Note:
- [Note] Detected use of non-constant indexes: z[1, 1, 1], z[1, 2, 1], z[1, 3, 1], z[1, 4, 1], z[1, 5, 1], ... For computational efficiency, we recommend specifying these in 'constants'.
I would appreciate any insights into why these errors may be occurring and how I can resolve them. I have attached the relevant model code for your reference.
# write nimble model code
code <- nimbleCode({
# intercepts
for (s in 1:3) {
phi_alpha[s] ~ dbeta(1, 1)
p_alpha[s] ~ dbeta(1, 1)
}
# sex effects
phi_beta ~ dnorm(0, 1)
p_beta ~ dnorm(0, 1)
# random effect SDs
for (k in 1:2) {
phi_sigma[k] ~ dexp(1)
p_sigma[k] ~ dexp(1)
}
# random year effects
for (t in 1:max(n_year)) {
phi_year_z[t] ~ dnorm(0, 1)
phi_year[t] <- phi_year_z[t] * phi_sigma[2]
p_year_z[t] ~ dnorm(0, 1)
p_year[t] <- p_year_z[t] * p_sigma[2]
}
# probability of transitioning to adulthood
tau_alpha ~ dbeta(1, 1)
# probability of being female for unobserved individuals
pr_female ~ dbeta(5, 5)
# open site loop
for (m in 1:n_site) {
# random site effects
phi_site_z[m] ~ dnorm(0, 1)
phi_site[m] <- phi_site_z[m] * phi_sigma[1]
p_site_z[m] ~ dnorm(0, 1)
p_site[m] <- p_site_z[m] * p_sigma[1]
# open individual loop
for (i in 1:n_ind[m]) {
# sex (being female) imputation
female[m, i] ~ dbern(pr_female)
# open years loop
for (t in first[m, i]:(n_year[m] - 1)) {
# survival probabilities
for (s in 1:3) {
phi[s, m, i, t] <- expit(phi_alpha[s] # age-specific intercept
+ phi_beta * (female[m, i] - 0.5) # sex effect (specified at -0.5, 0.5 for male/female)
+ phi_site[m] # random site effects
+ phi_year[t] # random year effects
)
} # s
# transition probability
# note: this is probabably a "nuisance parameter"; it doesn't tell you much, because a tortoise MUST transition to adulthood at some point which is more a function of its survival
tau[m, i, t] <- tau_alpha # intercept only
# ecological process
# departing state: <1 year
ps[1, m, i, t, 1] <- 0 # survives as <1, doesn't transition to juvenile (latter impossible)
ps[1, m, i, t, 2] <- phi[1, m, i,t] # survives as <1, MUST transition to juvenile!
ps[1, m, i, t, 3] <- 0 # survives as <1, transitions to adult (impossible)
ps[1, m, i, t, 4] <- 1 - phi[1, m, i, t] # doesn't survive as <1 (and thus didn't transition to juvenile)
# departing state: juvenile
ps[2, m, i, t, 1] <- 0 # transitions to <1 (impossible)
ps[2, m, i, t, 2] <- phi[2, m, i, t] * (1 - tau[m, i, t]) # survives as juvenile, doesn't transition to adult
ps[2, m, i, t, 3] <- phi[2, m, i, t] * tau[m, i, t] # survives as juvenile, transitions to adult
ps[2, m, i, t, 4] <- 1 - phi[2, m, i, t] # doesn't survive as juvenile
# departing state: adult
ps[3, m, i, t, 1] <- 0 # transitions to <1 (impossible)
ps[3, m, i, t, 2] <- 0 # transitions to juvenile (impossible)
ps[3, m, i, t, 3] <- phi[3, m, i, t] # survives as adult
ps[3, m, i, t, 4] <- 1 - phi[3, m, i, t] # doesn't survive as adult
# departing state: dead
ps[4, m, i, t, 1] <- 0
ps[4, m, i, t, 2] <- 0
ps[4, m, i, t, 3] <- 0
ps[4, m, i, t, 4] <- 1
# detection probabilities
for (s in 1:3) {
p[s, m, i, t] <- expit(p_alpha[s]
+ p_beta * (female[m, i] - 0.5) # sex effect (specified at -0.5, 0.5 for male/female)
+ p_site[m] # random site effects
+ p_year[t] # random year effects
)
} # s
# observation process
# note: this can be inside of the same time loop
# departing state: <1 year
po[1, m, i, t, 1] <- p[1, m, i, t]
po[1, m, i, t, 2] <- 0
po[1, m, i, t, 3] <- 0
po[1, m, i, t, 4] <- 1 - p[1, m, i, t]
# departing state: juvenile
po[2, m, i, t, 1] <- 0
po[2, m, i, t, 2] <- p[2, m, i, t]
po[2, m, i, t, 3] <- 0
po[2, m, i, t, 4] <- 1 - p[2, m, i, t]
# departing state: adult
po[3, m, i, t, 1] <- 0
po[3, m, i, t, 2] <- 0
po[3, m, i, t, 3] <- p[3, m, i, t]
po[3, m, i, t, 4] <- 1 - p[3, m, i, t]
# departing state: dead
po[4, m, i, t, 1] <- 0
po[4, m, i, t, 2] <- 0
po[4, m, i, t, 3] <- 0
po[4, m, i, t, 4] <- 1
} # t
# likelihood
for (t in (first[m, i] + 1):n_year[m]) {
# ecological process
z[m, i, t] ~ dcat(ps[z[m, i, t - 1], m, i, t - 1, 1:4])
# observation process
y[m, i, t] ~ dcat(po[z[m, i, t], m, i, t - 1, 1:4])
} # t
} # i
# derived quantities
# site- and age-specific survival and recapture
phi_m[m, 1:3] <- expit(phi_alpha[1:3] + phi_site[m])
p_m[m, 1:3] <- expit(p_alpha[1:3] + p_site[m])
# site- and sex-specific survival and recapture
phi_female[m] <- expit(phi_alpha[3] + phi_site[m] + phi_beta * 0.5)
phi_male[m] <- expit(phi_alpha[3] + phi_site[m] + phi_beta * -0.5)
p_female[m] <- expit(p_alpha[3] + p_site[m] + p_beta * 0.5)
p_male[m] <- expit(p_alpha[3] + p_site[m] + p_beta * -0.5)
} # m
# year-specific survival and recapture for adults
for (t in 1:(max(n_year) - 1)) {
phi_t[t] <- expit(phi_alpha[3] + mean(phi_site[1:n_site]) + phi_year[t])
p_t[t] <- expit(p_alpha[3] + mean(p_site[1:n_site]) + p_year[t])
}
})
# constants, data, initial values, and parameters to monitor
consts <- list(n_site = n_site,
n_ind = n_ind,
n_year = years$n_year,
first = first) |>
glimpse()
data <- list(y = ch,
z = z_known,
female = female) |>
glimpse()
inits <- list(z = z_init) |>
glimpse()
params <- c("alpha", "beta", "sigma", "site", "year", "m", "t", "female", "male")
mons <- c(str_c("phi_", params),
str_c("p_", params),
"tau_alpha",
"pr_female") # updated parameter name
# model
Rmodel <- nimbleModel(code, consts, data, inits, check = F, calculate = F)
Thank you for your assistance!
Best regards,
Motwaly
Chris Paciorek
--
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 on the web visit https://groups.google.com/d/msgid/nimble-users/5f6c0946-da22-4ceb-811f-23badbe32438n%40googlegroups.com.
B Motwaly
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/CA%2B7Ts_p6-twZ%3Df0%3DgoBa5F9%3Dk0z4pLJmhH97%3D6KGm6DAQXv2Mw%40mail.gmail.com.
Chris Paciorek
```
You'll want to check that for all the {m,i} pairs that that indexing produces a valid, non-empty, increasing sequence given your values of `first` and `n_year`.