LmW*_*mW. 5 r dplyr tidyr tidyverse
用tidyverse中的列表列数据结构拟合随数据帧的行而变化的不同模型公式的最佳方法是什么?
在R for Data Science中,Hadley给出了一个很棒的示例,说明了如何使用列表列数据结构和轻松地拟合许多模型(http://r4ds.had.co.nz/many-models.html#gapminder)。我正在尝试找到一种方法来适合具有稍微不同的公式的许多模型。在下面的示例中(改编自他的原始示例),为每个洲拟合不同模型的最佳方法是什么?
library(gapminder)
library(dplyr)
library(tidyr)
library(purrr)
library(broom)
by_continent <- gapminder %>%
group_by(continent) %>%
nest()
by_continent <- by_continent %>%
mutate(model = map(data, ~lm(lifeExp ~ year, data = .)))
by_continent %>%
mutate(glance=map(model, glance)) %>%
unnest(glance, .drop=T)
## A tibble: 5 × 12
# continent r.squared adj.r.squared sigma statistic p.value df
# <fctr> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
#1 Asia 0.4356350 0.4342026 8.9244419 304.1298 6.922751e-51 2
#2 Europe 0.4984659 0.4970649 3.8530964 355.8099 1.344184e-55 2
#3 Africa 0.2987543 0.2976269 7.6685811 264.9929 6.780085e-50 2
#4 Americas 0.4626467 0.4608435 6.8618439 256.5699 4.354220e-42 2
#5 Oceania 0.9540678 0.9519800 0.8317499 456.9671 3.299327e-16 2
## ... with 5 more variables: logLik <dbl>, AIC <dbl>, BIC <dbl>,
## deviance <dbl>, df.residual <int>
Run Code Online (Sandbox Code Playgroud)
我知道我可以通过遍历by_continent来做到这一点(效率不高,因为它估计了每个大陆的每个模型:
formulae <- list(
Asia=~lm(lifeExp ~ year, data = .),
Europe=~lm(lifeExp ~ year + pop, data = .),
Africa=~lm(lifeExp ~ year + gdpPercap, data = .),
Americas=~lm(lifeExp ~ year - 1, data = .),
Oceania=~lm(lifeExp ~ year + pop + gdpPercap, data = .)
)
for (i in 1:nrow(by_continent)) {
by_continent$model[[i]] <- map(by_continent$data, formulae[[i]])[[i]]
}
by_continent %>%
mutate(glance=map(model, glance)) %>%
unnest(glance, .drop=T)
## A tibble: 5 × 12
# continent r.squared adj.r.squared sigma statistic p.value df
# <fctr> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
#1 Asia 0.4356350 0.4342026 8.9244419 304.1298 6.922751e-51 2
#2 Europe 0.4984677 0.4956580 3.8584819 177.4093 3.186760e-54 3
#3 Africa 0.4160797 0.4141991 7.0033542 221.2506 2.836552e-73 3
#4 Americas 0.9812082 0.9811453 8.9703814 15612.1901 4.227928e-260 1
#5 Oceania 0.9733268 0.9693258 0.6647653 243.2719 6.662577e-16 4
## ... with 5 more variables: logLik <dbl>, AIC <dbl>, BIC <dbl>,
## deviance <dbl>, df.residual <int>
Run Code Online (Sandbox Code Playgroud)
但是是否有可能做到这一点而无需在基准R中返回循环(并且避免了我不需要的拟合模型)?
我试过的是这样的:
by_continent <- by_continent %>%
left_join(tibble::enframe(formulae, name="continent", value="formula"))
by_continent %>%
mutate(model=map2(data, formula, est_model))
Run Code Online (Sandbox Code Playgroud)
但是我似乎无法提出一个有效的est_model函数。我尝试了此功能(h / t:https : //gist.github.com/multidis/8138757),该功能不起作用:
est_model <- function(data, formula, ...) {
mc <- match.call()
m <- match(c("formula","data"), names(mc), 0L)
mf <- mc[c(1L, m)]
mf[[1L]] <- as.name("model.frame")
mf <- eval(mf, parent.frame())
data.st <- data.frame(mf)
return(data.st)
}
Run Code Online (Sandbox Code Playgroud)
(诚然,这是一个人为的例子。我的实际情况是,我有大量观测值缺失了数据中的关键自变量,因此我想将一个模型与完整观测值中的所有变量拟合,将另一个模型拟合至模型中的所有变量。休息观察。)
更新
我想出了一个有效的est_model函数(尽管可能效率不高):
est_model <- function(data, formula, ...) {
map(list(data), formula, ...)[[1]]
}
by_continent <- by_continent %>%
mutate(model=map2(data, formula, est_model))
by_continent %>%
mutate(glance=map(model, glance)) %>%
unnest(glance, .drop=T)
## A tibble: 5 × 12
# continent r.squared adj.r.squared sigma statistic p.value df
# <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
#1 Asia 0.4356350 0.4342026 8.9244419 304.1298 6.922751e-51 2
#2 Europe 0.4984677 0.4956580 3.8584819 177.4093 3.186760e-54 3
#3 Africa 0.4160797 0.4141991 7.0033542 221.2506 2.836552e-73 3
#4 Americas 0.9812082 0.9811453 8.9703814 15612.1901 4.227928e-260 1
#5 Oceania 0.9733268 0.9693258 0.6647653 243.2719 6.662577e-16 4
## ... with 5 more variables: logLik <dbl>, AIC <dbl>, BIC <dbl>, deviance <dbl>,
## df.residual <int>
Run Code Online (Sandbox Code Playgroud)
我发现列出模型公式更容易。每个模型只适合相应的一次continent
。我formula
向嵌套数据添加一个新列,以确保 和formula
的continent
顺序相同,以防万一它们不是。
formulae <- c(\n Asia= lifeExp ~ year,\n Europe= lifeExp ~ year + pop,\n Africa= lifeExp ~ year + gdpPercap,\n Americas= lifeExp ~ year - 1,\n Oceania= lifeExp ~ year + pop + gdpPercap\n)\n\ndf <- gapminder %>%\n group_by(continent) %>%\n nest() %>%\n mutate(formula = formulae[as.character(continent)]) %>%\n mutate(model = map2(formula, data, ~ lm(.x, .y))) %>%\n mutate(glance=map(model, glance)) %>%\n unnest(glance, .drop=T)\n\n# # A tibble: 5 \xc3\x97 12\n# continent r.squared adj.r.squared sigma statistic p.value df logLik AIC BIC\n# <fctr> <dbl> <dbl> <dbl> <dbl> <dbl> <int> <dbl> <dbl> <dbl>\n# 1 Asia 0.4356350 0.4342026 8.9244419 304.1298 6.922751e-51 2 -1427.65947 2861.31893 2873.26317\n# 2 Europe 0.4984677 0.4956580 3.8584819 177.4093 3.186760e-54 3 -995.41016 1998.82033 2014.36475\n# 3 Africa 0.4160797 0.4141991 7.0033542 221.2506 2.836552e-73 3 -2098.46089 4204.92179 4222.66639\n# 4 Americas 0.9812082 0.9811453 8.9703814 15612.1901 4.227928e-260 1 -1083.35918 2170.71836 2178.12593\n# 5 Oceania 0.9733268 0.9693258 0.6647653 243.2719 6.662577e-16 4 -22.06696 54.13392 60.02419\n# # ... with 2 more variables: deviance <dbl>, df.residual <int>\n
Run Code Online (Sandbox Code Playgroud)\n