Hi,
The issue maybe is that, my times are not really 3 observations per subject, as defined in your data. My data is unbalanced. I have also changed the inits function, so that it would not turn out to inf or nan, but I still get the same error message. How does your response from the generated data look like?
When I run that code, with the following sample function;
slanat ← cmdstan_model(“C:/Users/IGarciaFogeda/Documents/WP4/DRC/code/Mechanistic models/andraudexample.stan”)
fit_slanat ← slanat$sample(
seed = 123,
chains = 1,
iter_warmup = 1, iter_sampling = 1, step_size = 0.05,
refresh = 500 # print update every 500 iters
)
I get the following error message:
That’s not a fatal error, it’s a warning. If the warnings stop after the first few hundred iterations then it’s not a problem.
You could set init= and an even smaller step_size= and see if that prevents the warning. And don’t put a lower bound on logAB0.
I mentioned earlier that I added this right before the lognormal
if (is_nan(new_mu[i,3])) {
print("t = ", ts[i], "; inits = ", inits, "; theta = ", theta);
print("y = ", y[i], "; mu = ", mu[i,3], "; new_mu = ", new_mu[i,3]);
}
(That’s not going to fix anything but it’ll show you what the model was doing right before failing.)
Hi,
I have been running different models since I realised that my data was not sorted per subject.
I have tried the following things:
Sorting times per subject (0, 22, 70, 0, 50, 0, 30…) or sorting the times from the minimum value (several 0) to the maximum. With both of the options I get the following warning, which triggers the model to not converge, thus the Rhat goes beyond 1.
Lognormal_lpdf: location parameter is inf, but must be finite! line 109
The same occurs if I run the analytical model that we have mentioned previously in this post.
functions {
real[] sldecay(real t, real[] y, real[] theta, real[] x_r, int[] x_i) {
real dydt[3];
dydt[1] = -theta[1]*y[1];
dydt[2] = -theta[2]*y[2];
dydt[3] = theta[1]*y[1] + theta[2]*y[2] - theta[3]*y[3];
return dydt;
}
}
data {
int<lower=1> N;
int<lower=1> nsubjects;
int<lower=1> subj_ids[N];
real <lower=0> y[N];
int <lower=1> ntimes;
real ts[ntimes];
real t0;
}
transformed data {
real x_r[0];
int x_i[0];
}
parameters {
real logps;
real logpl;
real logmuab;
real sigmalogmuab;
real <lower= 0> logAB0;
real sigmalogAB0;
real logps0;
real logpl0;
real sigma;
vector[2] r[nsubjects];
//real rho_rhobit;
}
model{
real mu[1, 3];
real inits[3];
real theta[3];
real new_mu[ntimes, 3];
vector[nsubjects] rAB0;
vector[nsubjects] rmuab;
real eval_time[1];
// transformed parameters
real ps = exp(logps);
real pl = exp(logpl);
real muab = exp(logmuab);
real sigma_muab = exp(sigmalogmuab);
real AB0 = exp(logAB0);
real sigma_AB0 = exp(sigmalogAB0);
real ps0 = exp(logps0);
real pl0 = exp(logpl0);
real sigmap = exp(sigma);
// defining the correlation
vector[2] mean_vec;
matrix[2,2] Sigma;
//real rho = (exp(rho_rhobit) - 1)/(exp(rho_rhobit) + 1);
//prior distributions
logps ~ normal(0, 1);
logpl ~ normal(0, 1);
logmuab ~ normal(0, 10);
sigmalogmuab ~ normal(0, 1);
logAB0 ~ normal(3, .5);
sigmalogAB0 ~ normal(0, 1);
logps0 ~ normal(0, 1);
logpl0 ~ normal(0, 1);
sigma ~ normal(0, 1);
//rho_rhobit ~ uniform(-10, 10);
//cov matrix
mean_vec[1] = -pow(sigma_AB0, 2.0)/2;
mean_vec[2] = -pow(sigma_muab, 2.0)/2;
Sigma[1,1] = pow(sigma_AB0, 2.0);
Sigma[2,2] = pow(sigma_muab, 2.0);
Sigma[1,2] = 0;
Sigma[2,1] = 0;
for (j in 1:nsubjects) {
r[j] ~ multi_normal_cholesky(mean_vec, Sigma);
rAB0[j] = r[j,1];
rmuab[j] = r[j,2];
}
for (i in 1:N){
//defining the initial conditions
inits[3] = AB0*exp(rAB0[subj_ids[i]]);
inits[1] = inits[3]/ps0;
inits[2] = inits[3]/pl0;
//defining the parameters
theta[1] = ps;
theta[2] = pl;
theta[3] = muab*exp(rmuab[subj_ids[i]]);
// ode solver
eval_time[1] = ts[i];
if (eval_time[1] == 0){
new_mu[i,1] = inits[1];
new_mu[i,2] = inits[2];
new_mu[i,3] = log(inits[3]) - pow(sigmap, 2.0)/2;}
else{
mu = integrate_ode_rk45(sldecay, inits, t0, eval_time, theta, x_r, x_i);
new_mu[i,1] = mu[1,1];
new_mu[i,2] = mu[1,2];
new_mu[i,3] = log(mu[1,3]) - pow(sigma, 2.0)/2;}
y[i] ~ lognormal(new_mu[i, 3], sigmap);
}
}
Cholesky factor is like a square root. Don’t square the sigmas if you use multi_normal_cholesky
Sigma[1,1] = sigma_AB0;
Sigma[2,2] = sigma_muab;
Was that supposed to be sigmap like in the other branch?
Constraining the priors for sigma parameters reduces divergencies substantially
sigmalogmuab ~ normal(0, 0.1);
sigmalogAB0 ~ normal(0, 0.1);
sigma ~ normal(0, 0.1);
Instead of log-transforming these you could declare them directly as constrained types
parameters {
real<lower=0> sigma_muab;
real<lower=0> sigma_AB0;
}
model {
sigma_muab ~ exponential(1.0);
sigma_AB0 ~ exponential(1.0);
}
I don’t think logAB0 needs a <lower=0>. The implicit constraining transform is may cause problems there.
Hi,
You mean these parameters to be transformed with the exp(1)? But keep the specification of normal(0, 0.01) as a prior?
Sigmap before, it means the exp(logsigma), to restrict this to positive.
I mean, you don’t need to do
parameters {
real logsigma;
} transformed parameters {
real sigma = exp(logsigma);
just to restrict to positive values because
parameters {
real<lower=0> sigma;
}
works just as well.
I think you need to consider more carefully what kind of priors you want these parameters to have. exponential(1.0) was just one example that improves r-hats but isn’t super narrow.
Hi,
I was dealing with issues of convergence as well by trying out this.
I have posted the codes in another post, what could be causing then the non-convergence?