R的性能包中的check_model函数中的“视口为零维度”是什么意思？

Answer 1

在check_model 中有check_model.default。

check_model.default <- function(x,
                                dot_size = 2,
                                line_size = .8,
                                panel = TRUE,
                                check = "all",
                                alpha = .2,
                                dot_alpha = .8,
                                colors = c("#3aaf85", "#1b6ca8", "#cd201f"),
                                theme = "see::theme_lucid",
                                detrend = FALSE,
                                verbose = TRUE,
                                ...) {
  # check model formula
  if (verbose) {
    insight::formula_ok(x)
  }
  
  minfo <- insight::model_info(x, verbose = FALSE)
  
  if (minfo$is_bayesian) {
    ca <- suppressWarnings(.check_assumptions_stan(x))
  } else if (minfo$is_linear) {
    ca <- suppressWarnings(.check_assumptions_linear(x, minfo))
  } else {
    ca <- suppressWarnings(.check_assumptions_glm(x, minfo))
  }
  # else {
  #   stop(paste0("`check_assumptions()` not implemented for models of class '", class(x)[1], "' yet."), call. = FALSE)
  # }
  
  attr(ca, "panel") <- panel
  attr(ca, "dot_size") <- dot_size
  attr(ca, "line_size") <- line_size
  attr(ca, "check") <- check
  attr(ca, "alpha") <- alpha
  attr(ca, "dot_alpha") <- dot_alpha
  attr(ca, "detrend") <- detrend
  attr(ca, "colors") <- colors
  attr(ca, "theme") <- theme
  attr(ca, "model_info") <- minfo
  attr(ca, "overdisp_type") <- list(...)$plot_type
  ca
}

对于您的模型，首先在.check_assumptions_linear 中发生错误。

.check_assumptions_linear <- function(model, model_info) {
  dat <- list()
  
  dat$VIF <- .diag_vif(model)
  dat$QQ <- .diag_qq(model)
  dat$REQQ <- .diag_reqq(model, level = .95, model_info = model_info)
  dat$NORM <- .diag_norm(model)
  dat$NCV <- .diag_ncv(model)
  dat$HOMOGENEITY <- .diag_homogeneity(model)
  dat$OUTLIERS <- check_outliers(model, method = "cook")
  if (!is.null(dat$OUTLIERS)) {
    threshold <- attributes(dat$OUTLIERS)$threshold$cook
  } else {
    threshold <- NULL
  }
  dat$INFLUENTIAL <- .influential_obs(model, threshold = threshold)
  dat$PP_CHECK <- tryCatch(check_predictions(model), error = function(e) NULL)
  
  dat <- insight::compact_list(dat)
  class(dat) <- c("check_model", "see_check_model")
  dat
}

在.check_assumptions_linear 中，.diag_vif 中出现错误。似乎它适用于其他.diag_...s。

.diag_vif <- function(model) {
  out <- check_collinearity(model)
  dat <- insight::compact_list(out)
  if (is.null(dat)) {
    return(NULL)
  }
  dat$group <- "low"
  dat$group[dat$VIF >= 5 & dat$VIF < 10] <- "moderate"
  dat$group[dat$VIF >= 10] <- "high"
  
  dat <- datawizard::data_rename(
    dat,
    c("Term", "VIF", "SE_factor", "Component"),
    c("x", "y", "se", "facet")
  )
  
  dat <- datawizard::data_select(dat, c("x", "y", "facet", "group"))
  
  if (insight::n_unique(dat$facet) <= 1) {
    dat$facet <- NULL
  }
  
  attr(dat, "CI") <- attributes(out)$CI
  dat
}

在.diag_vif 中，datawizard::data_rename 部分出现错误。

它工作到

out <- check_collinearity(model)
dat <- insight::compact_list(out)
if (is.null(dat)) {
  return(NULL)
}
dat$group <- "low"
dat$group[dat$VIF >= 5 & dat$VIF < 10] <- "moderate"
dat$group[dat$VIF >= 10] <- "high"

其中dat 是

Low Correlation

        Term  VIF      VIF  CI Increased SE Tolerance Tolerance  CI group
 Coffee_Cups 1.04 [1.00, 1.85]         1.02      0.96  [0.54, 1.00]   low
  Start_Work 1.04 [1.00, 1.85]         1.02      0.96  [0.54, 1.00]   low
Warning message:
In length(other == 1) && paste0(other, "_CI") %in% colnames(x) :
  'length(x) = 2 > 1' in coercion to 'logical(1)'

然后错误

dat <- datawizard::data_rename(
  dat,
  c("Term", "VIF", "SE_factor", "Component"),
  c("x", "y", "se", "facet")
)

Error in `colnames<-`(`*tmp*`, value = `*vtmp*`) : 
  attempt to set 'colnames' on an object with less than two dimensions

奇怪的是，如果我们检查names(dat)，我们可以看到

[1] "Term"              "VIF"               "VIF_CI_low"        "VIF_CI_high"      
[5] "SE_factor"         "Tolerance"         "Tolerance_CI_low"  "Tolerance_CI_high"
[9] "group"  

但是当我们查看dat时没有"SE_factor", "Component"

我们可以在datawizard::data_rename 中使用as.data.frame(dat) 而不是dat 来消除这个错误

然后，下一个错误发生在

class(dat) <- c("check_model", "see_check_model")

如果我们从.check_assumptions_linear 中删除这部分并尝试check_model.default(fit.work)，该函数会给出正确的结果，用它的其他组件绘制$PP_CHECK 的东西。

您可能会在以下位置找到它的源代码

check_model, check_model_diagostics 和 check_outliers

Answer 2

我不确定最后一个答案中的补救措施是什么，所以我找到了一个至少适合我需要的简单解决方法。首先，我将check_model 保存为对象：

#### Save Checks Into Object ####
check <- check_model(fit.work)

要手动提取数据，我只是使用普通的$ 运算符来提取变量：

#### Manually Check Data ####
check$OUTLIERS

例如：

OK: No outliers detected.

然后使用基本 R plot 函数手动绘制它们，如下所示：

#### Manually Plot ####
plot(check$OUTLIERS)
plot(check$PP_CHECK)