How to get model parameter summary of selected variables

General
1

The summary function returns the describe of all model variables. Anyway, it takes a long time and I only need the summary of model parameter variables rather than latent variables and generated quantities variables. However , I’ve searched the manual paper but dont get the wanted API function。
Here is the example

// Bayesian Normal Linear Regression Model (IMOD)
data {
    int<lower=0> N;                 // Number of observations
    vector[N] w;                    // Regressor
    vector[N] y;                    // Response variable
}

parameters {
    real alpha;                     // Intercept
    real beta;                      // Slope
    real<lower=0> scaled_h;               // Scale parameter
}

transformed parameters {
    real<lower=0> h = 1/3 * scaled_h; // 进行缩放
}

model {
    // Priors
    alpha ~ normal(0, 10);         // Normal prior for alpha
    beta ~ normal(0, 10);          // Normal prior for beta
    scaled_h ~ chi_square(3);              // Chi-square prior for h 

    // Likelihood  method 1

    //for (i in 1:N) {
    //    y[i] ~ normal(alpha + beta * w[i], 1 / sqrt(scaled_h)); // y_i given w_i
   // }

    // Likelihood  method 2
        y ~ normal(alpha + beta * w, 1 / sqrt(scaled_h)); // y_i given w_i


}

generated quantities {
    vector[N] log_like;  // 初始化 log_like 数组
    vector[N] st_alpha;   // 关于 alpha 的导数
    vector[N] st_beta;    // 关于 beta 的导数
    vector[N] st_h; // 关于 h 的导数

    for (i in 1:N) {
        real mu = alpha + beta * w[i]; // 预测值
        real scale = 1 / sqrt(h); // 标准差

        log_like[i] = normal_lpdf(y[i] | mu, scale); // 计算每个观察值的对数似然

        // 计算导数
        st_alpha[i] = - (y[i] - mu) * scale; // 关于 alpha 的导数
        st_beta[i] = - (y[i] - mu) * w[i] * scale; // 关于 beta 的导数
        st_h[i] = -0.5 * (y[i] - mu) * (y[i] - mu) - 0.5 / h; // 关于 h 的导数
    }

    // 将导数拼接到一个变量中
    matrix[N, 3] st; 
    for (i in 1:N) {
        st[i, 1] = st_alpha[i];
        st[i, 2] = st_beta[i];
        st[i, 3] = st_h[i]; 
    }
}

编译模型

model = cmdstanpy.CmdStanModel(stan_file='bayes_normal.stan')

运行采样

fit = model.sample(data = data_1,chains=4)

# 查看拟合结果的摘要信息

# print(fit.summary())
print(fit.summary(fit.stan_variable('beta'，'h'))

The last sentence is my purpose but it doesnot follow the rule. And I wonder how to tackle it.
Moreover, is there any group in any apps we can chat about the questions of cmdstan with each other.

[edit: escaped code]

2

Hi, @tianyi_liu, and welcome to the Stan forums. Sorry nobody answered this relatively simple question.

I don’t think there’s a way in CmdStanPy to only get a summary for specific variables because it’s just passing the analysis through to CmdStan, which doesn’t give you a way to do this.

If you don’t want to see the generated quantities, remove the block. You can put it back in to compute them later using the standalone generated quantities functionality:

https://mc-stan.org/cmdstanpy/users-guide/examples/Run%20Generated%20Quantities.html

Also, you probably want log_lik instead of log_like if you are going to pass these into something like LOO in the ArviZ package.