Feedback request on updating depreciated model and addressing 'variable does not exist' error

Modeling
,
1

This RStan model was originally used in a publication from 2016 and I am trying to implement it in Pystan. I have done my best to update the code with respect to Stan’s current documentation, but I would like some feedback because I continue to get this error:

Exception: variable does not exist; processing stage=data initialization; variable name=f; base type=int (in ‘unknown file name’ at line 3)

data {
    int<lower=0> N; // Number of dates in the set
    int<lower=0> f[N]; // Number of failed components in a set
    int<lower=0> s[N]; // Number of non-failed components in a set
    real<lower=0> t[N]; // Time inspected 
    int<lower=2> p; // Number of covariates (intercept included)
    row_vector[p] X[N]; // Matrix [N,p] of covariates (intercept included)
    vector[p] a0; // Mean vector g-prior for alpha 
    vector[p] b0; // Mean vector g-prior for beta
    cov_matrix[p] V1; // Covariate matrix g-prior for alpha 
    cov_matrix[p] V2; // Covariate matrix g-prior for beta
}

parameters {
    vector[p] alpha; // parameter vector alpha
    vector[p] beta; // parameter vector beta
}

model {
    alpha ~ multi_normal(a0, V1); // gaussian-prior for alpha 
    beta ~ multi_normal(b0, V2); // gaussian-prior for beta
    
    for (i in 1:N) { 
        // log-likelihood (for right and left censored data)
        target += s[i]*weibull_lccdf(t[i] | exp(X[i]*alpha), exp(X[i]*beta));
        target += f[i]*log(weibull_cdf(t[i], exp(X[i]*alpha), exp(X[i]*beta)));
    }
}
generated quantities {
    vector[N] eta; // scale parameter for each observation
    for (i in 1:N) {
        eta[i] = exp(X[i]*beta);
    }
}

Here is my data. I originally had ndarrays of int32 for f and s, but I converted them to lists to avoid compatibility issues with C++. inspected_times is a list of times as floats, D is a list of deaths as ints, C is a list of censored as ints, cov_X is my feature (covariate) matrix

data = {
    'N' : len(inspected_times), # Number of dates in the set
    'f[N]' : D, # Number of failed components in a set
    's[N]' : C, # Number of non-failed components in a set
    't[N]' : inspected_times, # Time inspected
    'p' : len(cov_X[0]), # Number of covariates (calendar, causes, intercept included)
    'X' : cov_X, # [N,p] of covariates (calendar, cause, Sintercept)
    'a0' : 0, # Mean vector g-prior for alpha 
    'b0' : 0, # Mean vector g-prior for beta
    'V1' : len(inspected_times)*invert_matrix(cov_X.T@cov_X), # Cov. matrix g-prior for alpha 
    'V2' : len(inspected_times)*invert_matrix(cov_X.T@cov_X), # Cov. matrix g-prior for beta
}
2

The model doesn’t contain any errors to my brief knowledge, but there is an issue with my data inputs.

PyStan users: You can use lists or ndarrays (these need to contain ints) for stan arrays in PyStan 2.19.

The solution is I needed to drop dimensions from the parameter names in my data dictionary. So f[N] --> f. Also, there was a small typo, since ‘a0’ and ‘b0’ should be vectors of length p.