PyStan报错TypeError：无法将Series转换为<int>类型求助

问题场景 #

在Jupyter Notebook中使用PyStan，通过nest_asyncio.apply()适配异步运行，传入整数类型NumPy数组时触发上述错误，但输入已显式为整数类型。可复现代码如下：

import numpy as np
import pandas as pd
import stan 
import nest_asyncio 
nest_asyncio.apply()

def sigmoid(x):
    return 1 / (1 + np.exp(-x))

treatment_slope = 0.2
test_slope = 0.15
intercept = 0.7

def link(X):
    theta = sigmoid(X[0] * treatment_slope + X[1] * test_slope + intercept)
    return theta

ctrl = np.random.binomial(n=1, p=link([0,0]), size=500)
test_no_exp = np.random.binomial(n=1, p=link([0,1]), size=250)
test_w_exp = np.random.binomial(n=1, p=link([0,1]), size=250)

data = {'test':[], 'exposure':[], 'outcome':[]}
for c in ctrl:
    data['test'].append(0)
    data['exposure'].append(0)
    data['outcome'].append(c)
    
for t in test_no_exp:
    data['test'].append(1)
    data['exposure'].append(0)
    data['outcome'].append(t)
    
for e in test_w_exp:
    data['test'].append(1)
    data['exposure'].append(1)
    data['outcome'].append(e)
    
df = pd.DataFrame(data)    
X = df[['test','exposure']].to_numpy()
y = df['outcome']

payload = {'X':X, 'y':y, 'K':1000, 'N':2}

logistic_regression_model = """
// logistic regression code
data {
    int<lower=0> N;                 // number of observations
    int<lower=0> K;                 // number of predictor variables
    matrix[N, K] X;                 // predictor matrix
    int<lower=0,upper=1> y[N];      // outcome vector
}

parameters {
    real alpha;                     // intercept
    vector[K] theta;                // coefficients on Q_ast
    real<lower=0> sigma;            // error scale
}  

model {
    // priors
    for (d in 1:D) {
        theta[d] ~ normal(0, 1);
    }
    alpha ~ normal(0, 10);
    // likelihood
    y ~ bernoulli_logit(X * theta + alpha);
}
"""
posterior = stan.build(logistic_regression_model, data=payload)
fit = posterior.sample(num_chains=4, num_samples=1000)

错误原因 #

1. N与K参数完全颠倒：Stan代码中N是观测样本数（此处应为1000），K是特征数（此处应为2），但payload里把两者赋值反了，导致矩阵维度不匹配，触发类型转换错误。
2. y的类型不匹配：y = df['outcome']得到的是Pandas Series对象，Stan要求传入整数类型的NumPy数组，直接传入Series会引发类型转换失败。
3. 未定义的循环变量D：Stan模型中for (d in 1:D)的D未在data块声明，实际应该用已定义的K来遍历系数。
4. 冗余的sigma参数：逻辑回归使用伯努利分布，不需要尺度参数sigma，该参数无意义且会增加模型冗余。

修正后的代码 #

import numpy as np
import pandas as pd
import stan 
import nest_asyncio 
nest_asyncio.apply()

def sigmoid(x):
    return 1 / (1 + np.exp(-x))

treatment_slope = 0.2
test_slope = 0.15
intercept = 0.7

def link(X):
    theta = sigmoid(X[0] * treatment_slope + X[1] * test_slope + intercept)
    return theta

ctrl = np.random.binomial(n=1, p=link([0,0]), size=500)
test_no_exp = np.random.binomial(n=1, p=link([0,1]), size=250)
test_w_exp = np.random.binomial(n=1, p=link([0,1]), size=250)

data = {'test':[], 'exposure':[], 'outcome':[]}
for c in ctrl:
    data['test'].append(0)
    data['exposure'].append(0)
    data['outcome'].append(c)
    
for t in test_no_exp:
    data['test'].append(1)
    data['exposure'].append(0)
    data['outcome'].append(t)
    
for e in test_w_exp:
    data['test'].append(1)
    data['exposure'].append(1)
    data['outcome'].append(e)
    
df = pd.DataFrame(data)    
X = df[['test','exposure']].to_numpy()
# 修正点1：将y转为NumPy整数数组
y = df['outcome'].to_numpy(dtype=int)

# 修正点2：颠倒N和K的赋值，匹配Stan定义
payload = {'X':X, 'y':y, 'K':2, 'N':1000}

logistic_regression_model = """
// logistic regression code
data {
    int<lower=0> N;                 // number of observations
    int<lower=0> K;                 // number of predictor variables
    matrix[N, K] X;                 // predictor matrix
    int<lower=0,upper=1> y[N];      // outcome vector
}

parameters {
    real alpha;                     // intercept
    vector[K] theta;                // coefficients on predictors
}  

model {
    // priors
    // 修正点3：用K替代未定义的D
    for (d in 1:K) {
        theta[d] ~ normal(0, 1);
    }
    alpha ~ normal(0, 10);
    // likelihood
    y ~ bernoulli_logit(X * theta + alpha);
}
"""
posterior = stan.build(logistic_regression_model, data=payload)
fit = posterior.sample(num_chains=4, num_samples=1000)

内容的提问来源于stack exchange，提问作者jbuddy_13