我是贝叶斯分析领域的超级新手,我正在尝试使用经典捕获-捕获模型Mh2的示例进行练习
这是我的代码
nind <- dim(venados)[1]
K <- 43
ntraps <- 13
M <- 150
nz <- M - nind
Yaug <- array(0, dim = c(M, ntraps, K))
Yaug[1:nind,,] <- venados
y <- apply(Yaug, c(1,3), sum)
y[y > 1] <- 1
Bundle data
data1 <- list(y = y, nz = nz, nind = nind, K = K, sup = Buffer)
# Model JAGS
sink("Mh2_jags.txt")
cat("
model{
# Priors
p0 ~ dunif(0,1)
mup <- log(p0/(1-p0))
sigmap ~ dunif(0,10)
taup <- 1/(sigmap*sigmap)
psi ~ dunif(0,1)
# Likelihood
for (i in 1:(nind+nz)) {
z[i] ~ dbern(psi)
lp[i] ~ dnorm(mup,taup)
logit(p[i]) <- lp[i]
y[i] ~ dbin(mu[i],K)
} # i
N <- sum(z[1:(nind+nz)])
D <- N/sup*100
} # modelo
",fill = TRUE)
sink()
# Inicial values
inits <- function(){list(z = as.numeric(y >= 1), psi = 0.6, p0 = runif(1), sigmap = runif(1, 0.7, 1.2), lp = rnorm(M, -0.2))}
params1 <- c("p0","sigmap","psi","N","D")
# MCMC
ni <- 10000; nt <- 1; nb <- 1000; nc <- 3
# JAGS and posteriors
fM2 <- jags(data1, inits, params1, "Mh2_jags.txt", n.chains = nc, n.thin = nt, n.iter = ni, n.burnin = nb)
我收到此错误消息
Processing function input.......
Done.
Compiling model graph
Resolving undeclared variables
Deleting model
Error in jags.model(file = model.file, data = data, inits = inits, n.chains = n.chains, :
RUNTIME ERROR:
Compilation error on line 16.
Dimension mismatch in subset expression of y
我读到一些字母,如s和n必须更改。然而, 我不知道该怎么办。请提供意见。
非常感谢您
答案 0 :(得分:0)
问题是因为y
是二维的,但是模型假定它是一维的。如果您假设次要调查是i.i.d. Bernoulli试用(每个会话都有K
个试用)n,那么您只需要取y
矩阵的行之和即可。假设是这种情况,那么您只需要在此脚本的顶部修改几行即可。
nind <- dim(venados)[1]
K <- 43
ntraps <- 13
M <- 150
nz <- M - nind
Yaug <- array(0, dim = c(M, ntraps, K))
Yaug[1:nind,,] <- venados
y <- apply(Yaug, c(1,3), sum)
y[y > 1] <- 1
# Take the rowSum
y_vector <- rowSums(y)
# Use y_vector instead of y
data1 <- list(y = y_vector, nz = nz, nind = nind, K = K, sup = Buffer)
相反,如果要包括观测过程的协变量(并且这些协变量因调查而异),则可以使用矩阵y
并修改模型。
sink("Mh2_jags_Kloop.txt")
cat("
model{
# Priors
p0 ~ dunif(0,1)
mup <- log(p0/(1-p0))
sigmap ~ dunif(0,10)
taup <- 1/(sigmap*sigmap)
psi ~ dunif(0,1)
# Likelihood
for (i in 1:(nind+nz)) {
z[i] ~ dbern(psi)
lp[i] ~ dnorm(mup,taup)
logit(p[i]) <- lp[i]
# Loop over K surveys
for(j in 1:K){
y[i,j] ~ dbern(p[i]*z[i])
}
} # i
N <- sum(z[1:(nind+nz)])
D <- N/sup*100
} # modelo
",fill = TRUE)
sink()
最后,您没有指定模型中的mu
。我认为您希望它是p
,但是您还需要将潜在状态模型链接到观察状态模型(如果z=0
则无法对该人进行采样。在这种情况下,您可以解释{{ 1}}作为psi
个人在您网站上的概率。
nind+nz