如何在嵌套数据框中的模型上使用 MuMIn::dredge

Question

我在嵌套数据框中有许多 glmmTMB 模型，我想迭代地挖掘它们，但即使在尝试挖掘单个模型时我也会得到这个结果：

eval(call("is.data.frame", gmCall$data), gmEnv) 中的错误： 找不到对象“is.data.frame”

只要数据在环境中保存为对象，在嵌套数据框之外拟合相同的模型就不会出现问题。是否有修复使用嵌套数据框的问题？或者如果需要的话可以列出一份清单吗？我不想让数十个独立的数据框和随附的模型四处浮动。

使用

mtcars

数据集的示例：

# Example model to fit
mod_fx <- function(x) {
  glmmTMB(mpg ~ disp + drat + wt, 
          data = x,
          family = gaussian(),
          na.action = "na.fail")
}

# Grouping and nesting data
dat <- mtcars %>% 
  group_by(cyl) %>%
  nest() %>% 
  mutate(mod = map(data, mod_fx))

# Model summary works fine, but note: Data = x
# Related to problem?
summary(dat$mod[[1]])

# Dredge fails
dred_dat <- dredge(dat$mod[[1]],
                   rank = "AICc",
                   evaluate = T)

# Dredge does work when same model is built with separate data frame object stored in environment, not a nested data frame.
dat2 <- filter(mtcars, cyl == 6)

m1 <- glmmTMB(mpg ~ disp + drat + wt, 
        data = dat2,
        family = gaussian(),
        na.action = "na.fail")

summary(m1) # Same model results, (data = dat2), but dredge works:

dred_dat <- dredge(m1,
                   rank = "AICc",
                   evaluate = T)

# This also fails without data stored as separate object
m3 <- glmmTMB(mpg ~ disp + drat + wt, 
              data = filter(mtcars, cyl == 6),
              family = gaussian(),
              na.action = "na.fail")

summary(m3) # Data: filter(mtcars, cyl == 6)

dred_dat <- dredge(m3,
                   rank = "AICc",
                   evaluate = T)
# Error in dredge(m3, rank = "AICc", evaluate = T) : 'global.model' uses "data" that is a function value: use a variable instead

Answer 1

这里有一种可能性：将模型拟合和疏浚放在一个函数中，然后嵌套/映射。

library(tidyverse)
library(MuMIn)

dfun <- function(x) {
   g1 <- glmmTMB(mpg ~ disp + drat + wt, 
          data = x,
          family = gaussian(),
          na.action = "na.fail")
   dredge(g1, rank = "AICc",  evaluate = TRUE)
}
dat <- mtcars %>% 
  group_by(cyl) %>%
  nest() %>% 
  mutate(dredged = map(data, dfun))

 dat$dredged[[1]]
Global model call: glmmTMB(formula = mpg ~ disp + drat + wt, data = x, family = gaussian(), 
    na.action = "na.fail", ziformula = ~0, dispformula = ~1)
---
Model selection table 
  cnd((Int)) dsp((Int)) cnd(dsp) cnd(drt) cnd(wt) df  logLik AICc delta weight
1     19.740          +                            2 -12.011 31.0  0.00  0.750
5     28.410          +                    -2.780  3  -9.825 33.7  2.63  0.201
3     18.490          +            0.3503          3 -11.965 37.9  6.91  0.024
2     19.080          + 0.003605                   3 -11.974 37.9  6.93  0.024
6     28.190          + 0.019140           -3.835  4  -7.830 43.7 12.64  0.001
7     30.650          +           -0.4436  -2.990  4  -9.702 47.4 16.38  0.000
4      8.536          + 0.022440   1.9780          4 -11.481 51.0 19.94  0.000
8     16.050          + 0.042090   2.3460  -3.989  5  -4.630 79.3 48.24  0.000
Models ranked by AICc(x)