r - 按组汇总 wtd.quantile

Question

我想使用 Hmisc::wtd.quantile 为具有许多重复日期的数据框创建一个新的 df。我按日期分组，使用 summarize()按日期聚合，并尝试使用 wtd.quantile()在每个日期(带重量)。这也是一个相当大的数据集。下面是一些示例代码:

# sample data
# grouping_var = dt_time
require(Hmisc)
require(plyr)
require(dplyr)
df <- data.frame(type = sample(letters[1:2], 10e6, replace = TRUE), 
             score = sample(500:899, 10e6, replace = TRUE),
             dt_time = sample(seq(as.Date('2010/01/01'), 
                                  as.Date('2018/01/01'), 
                                  by="day"), 10e6, replace = TRUE),
             weight = sample(1.0:2.0, 10e6, replace = TRUE))
# my attempt:
ptiles <- df %>%
group_by(dt_time) %>%
plyr::ddply(~dt_time, dplyr::summarize,
            ptile10 = Hmisc::wtd.quantile(., .$score, weights = .$weight, 
probs = .1, na.rm = TRUE),
            ptile50 = Hmisc::wtd.quantile(., .$score, weights = .$weight, 
probs = .5, na.rm = TRUE),
            ptile90 = Hmisc::wtd.quantile(., .$score, weights = .$weight, 
probs = .9, na.rm = TRUE))

# desired df,
# where each new variable would be created using the
# wtd.quantile function:
desired_ptiles <- data.frame(dt_time = seq(as.Date('2010/01/01'),
                                       as.Date('2010/01/06'),
                                       by = "day"),
                         # only 6 because lol 10e6
                         ptile10 = sample(500:899, 6, replace = TRUE),
                         ptile50 = sample(500:899, 6, replace = TRUE),
                         ptile90 = sample(500:899, 6, replace = TRUE))

到目前为止，我的努力导致了这个错误:

Error in summarise_impl(.data, dots) :
Evaluation error: 'arg' must be NULL or a character vector.

当使用公式符号时:

ptiles <- df %>%
    group_by(dt_time) %>%
plyr::ddply(~dt_time, dplyr::summarize,
ptile10 = Hmisc::wtd.quantile(., .$score, weights = .$weight,
                                     probs = .1, na.rm = TRUE),
ptile50 = Hmisc::wtd.quantile(., .$score, weights = .$weight, 
                                        probs = .5, na.rm = TRUE),
          ptile90 = Hmisc::wtd.quantile(., .$score, weights = .$weight,
          probs = .9, na.rm = TRUE))
# error message:
Error in summarise_impl(.data, dots) : 
  Evaluation error: 'arg' must be NULL or a character vector.

我是否以错误的方式接近这个？我见过使用 split() 的方法但这似乎很烦人。有没有 data.table允许的方法 wtd.quantile()以这种方式概括？

谢谢!

Answer 1

你不需要ddply使用时 group_by ，因为数据已经被分组分割了。此外，您不需要在 summarize 中定义数据。分组后。

这有效:

ptiles <- df %>%
  group_by(dt_time) %>%
  summarize(ptile10 = wtd.quantile(score, weights = weight, 
                                            probs = .1, na.rm = TRUE),
              ptile50 = wtd.quantile(score, weights = weight, 
                                            probs = .5, na.rm = TRUE),
              ptile90 = wtd.quantile(score, weights = weight, 
                                            probs = .9, na.rm = TRUE))

> ptiles
# A tibble: 2,923 x 4
      dt_time ptile10 ptile50 ptile90
       <date>   <dbl>   <dbl>   <dbl>
 1 2010-01-01   539.0     697   859.0
 2 2010-01-02   538.0     704   861.7
 3 2010-01-03   541.0     706   862.0
 4 2010-01-04   541.0     702   859.0
 5 2010-01-05   540.0     706   860.0
 6 2010-01-06   537.0     695   859.0
 7 2010-01-07   539.0     696   859.0
 8 2010-01-08   536.0     700   857.0
 9 2010-01-09   538.0     694   861.0
10 2010-01-10   538.4     701   859.0
# ... with 2,913 more rows