When I use stan() in for sentences then sometimes, R hat and n_eff and se_mean does not be calculated even if such a chain is not constant. The following code use 6 fitted models and in the third model, R hat does not calculated. If the code does not reproduce this, then please chainge 6 or 3. It sometimes occurs and sometimes does not occur.
My original code is too bit to show, so, I made some simple example. (I guess, it still includes redundant stuff.) The code is the following. In this code, 6 fittted model objects (x[[i]], i= 1,..., 6) are created. In the third fitted model object, x[[3]], some R hat is not calculated as follows . In this example, a chains of such a prametere looks non-constant.
mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat
A[11] 0.71 0.02 0.16 0.39 0.64 0.72 0.82 0.98 65 0.99
A[12] 0.72 NaN 0.17 0.38 0.60 0.72 0.84 1.01 NaN NaN
A[13] 0.70 0.02 0.17 0.27 0.61 0.72 0.84 0.94 98 1.01
A[14] 0.71 0.01 0.12 0.42 0.65 0.73 0.79 0.89 98 1.00
A[15] 0.71 NaN 0.18 0.40 0.58 0.74 0.83 0.94 NaN NaN
A[16] 0.70 0.02 0.16 0.34 0.62 0.73 0.80 0.92 97 0.99
A[17] 0.71 NaN 0.20 0.27 0.58 0.71 0.86 1.08 NaN NaN
A[18] 0.71 NaN 0.15 0.40 0.61 0.73 0.82 0.95 NaN NaN
A[19] 0.73 0.01 0.12 0.51 0.64 0.74 0.82 0.93 98 1.00
A[20] 0.71 0.01 0.13 0.47 0.61 0.70 0.79 0.93 98 0.98
A[21] 0.72 0.01 0.13 0.46 0.65 0.73 0.81 0.91 98 1.00
A[22] 0.69 0.01 0.14 0.38 0.61 0.71 0.77 0.94 98 1.00
A[23] 0.70 0.01 0.13 0.45 0.62 0.70 0.79 0.94 98 0.98
A[24] 0.72 0.02 0.14 0.38 0.66 0.71 0.78 0.95 52 1.03
A[25] 0.69 0.01 0.14 0.37 0.63 0.69 0.77 0.95 98 0.99
A[26] 0.67 0.02 0.19 0.19 0.55 0.71 0.79 0.95 94 1.00
A[27] 0.71 0.02 0.17 0.39 0.60 0.74 0.84 0.99 91 0.99
A[28] 0.70 0.02 0.17 0.34 0.59 0.71 0.84 0.93 98 0.99
A[29] 0.67 0.01 0.14 0.34 0.58 0.69 0.78 0.91 98 0.98
A[30] 0.71 NaN 0.19 0.34 0.60 0.72 0.84 0.98 NaN NaN
lp__ -911.38 0.89 3.43 -918.23 -913.60 -911.20 -908.68 -905.89 15 1.12
But, when I show the trace plot of one of such a parameter, e.g., A[30], trace plot looks good. I am worried if the example does not replicate such a phenomenon.
Please Copy and paste the following R-code.
x <- list()
for (k in 1:6) {
stanmodelcode <- "
data {
int<lower=0> N;
real y[N];
}
parameters {
real mu[30];
}
model {
for(i in 1:30){
target += normal_lpdf(mu[i] | 0, 10);
target += normal_lpdf(y | mu[i], 1);
}
}
generated quantities{
real A[30];
for(i in 1:30){
A[i] = sqrt(mu[i]+1 );
}
}
"
y <- rnorm(20)
dat <- list(N = 20, y = y);
x[[k]] <- stan(model_code = stanmodelcode, model_name = "example",
data = dat, iter = 111, chains = 1, verbose = F )
}
x[[3]]
rstan::traceplot(x[[3]], pars=c("A[30]"))
