如何修复 R2jags::jags 中的“节点与父母不一致”

Posted 2023-02-19

【中文标题】如何修复 R2jags::jags 中的“节点与父母不一致”

【英文标题】：How to fix 'Node inconsistent with parents' in R2jags::jags

【发布时间】：2019-09-16 18:22:30

【问题描述】：

我正在使用 R-package R2jags。运行我下面附上的代码后，R 产生了错误消息：“节点与父节点不一致”。

我试图解决它。但是，错误消息仍然存在。我使用的变量是：

i) “采用”：一个 0-1 的虚拟变量。

ii) “NumInfo”：一个计数器变量，其范围是 0, 1, 2,...。

iii) “价格”：5

iv) “NRows”：326。

install.packages("R2jags")
library(R2jags)

# Data you need to run the model.
# Adop: a 0-1 dummy variable.
Adop <- c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)

# NumInfo: a counter variable.
NumInfo <- c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 2, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1)

# NRows: length of both 'NumInfo' and 'Adop'.
NRows <- length(NumInfo)

# Price: 5
Price <- 5

Data <- list("NRows" = NRows, "Adop" = Adop, "NumInfo" = NumInfo, "Price" = Price)

# The Bayesian model. The parameters I would like to infer are: 'mu.m', 'tau2.m', 'r.s', 'lambda.s', 'k', 'c', and 'Sig2'. 
# I would like to obtain samples from the posterior distribution of the vector of parameters.

Bayesian_Model <- "model 
    mu.m ~ dnorm(0, 1)                      
    tau2.m ~ dgamma(1, 1)           
    r.s ~ dgamma(1, 1)
    lambda.s ~ dgamma(1, 1)
    k ~ dunif(1, 1/Price)
    c ~ dgamma(1, 1)
    Sig2 ~ dgamma(1, 1)

    precision.m <- 1/tau2.m
    m ~ dnorm(mu.m, precision.m)
    s2 ~ dgamma(r.s, lambda.s)

    for(i in 1:NRows)
        Media[i] <- NumInfo[i]/Sig2 * m
        Var[i] <- equals(NumInfo[i], 0) * 10 + (1 - equals(NumInfo[i], 0)) * NumInfo[i]/Sig2 * s2 * (NumInfo[i]/Sig2 + 1/s2)
        Prec[i] <- pow(Var[i], -1)
        W[i] ~ dnorm(Media[i], Prec[i])
        PrAd1[i] <- 1 - step(-m/s2 - 1/c * 1/s2 * log(1 - k * Price) + 1/2 * c)
        PrAd2[i] <- 1 - step(-W[i] - m/s2 - 1/c * 1/s2 * log(1 - k * Price) + 1/2 * c - 1/c * log(1 - k * Price))
        PrAd[i] <- equals(NumInfo[i], 0) * PrAd1[i] + (1 - equals(NumInfo[i], 0)) * PrAd2[i]
        Adop[i] ~ dbern(PrAd[i])
        
    "

# Save the Bayesian model in your computer with an extension '.bug'.
# Suppose that you saved the .bug file in: "C:/Users/Default/Bayesian_Model.bug".
writeLines(Bayesian_Model, "C:/Users/Default/Bayesian_Model.bug")

# Here I would like to use jags command from R-package called R2jags.
# I would like to generate 1000 iterations.
MCMC_Bayesian_Model <- R2jags::jags(
    model.file = "C:/Users/Default/Bayesian_Model.bug",
    data = Data, 
    n.chains = 1, 
    n.iter = 1000,
    parameters.to.save = c("mu.m", "tau2.m", "r.s", "lambda.s", "k", "c", "Sig2")
    )

运行代码时，R 产生错误消息：“节点与父节点不一致”。我不知道错误是什么。我想知道你是否可以帮我解决这个问题，拜托。如果您需要更多信息，请告诉我。非常感谢。

【参考方案1】：

在不知道您要做什么的情况下找出模型有点困难，但我建议进行两个修复：

你指的是k ~ dunif(0, 1/Price)，而不是k ~ dunif(1, 1/Price)？对于dunif(a, b)，您必须拥有a < b（请参阅此处的第48页：http://people.stat.sc.edu/hansont/stat740/jags_user_manual.pdf）。

我在模型中插入了额外的一行，

PrAd01[i] <- max(min(PrAd[i], 0.99), 0.01)

并将最后一行更改为

Adop[i] ~ dbern(PrAd01[i])

上述手册的第 49 页指出 0 < p < 1 代表 dbern(p)。

模型运行上述两个变化。

【讨论】：

关于均匀分布，@WeihuangWong 先生您是对的。我在那里犯了一个错误。我听从了你的建议，它奏效了！！哟呼！！谢谢！