使用data.table计算运行金额不会计算没有值的天数？

Question

我有两个数据框，其中包含两个不同年份的值，按组分类，我想将它们组合在一起以计算每个组的运行金额，同时填写缺失的日期。这是我正在使用的示例数据框。

df1 <- data.frame(
  group = sample(c('g1', 'g2', 'g3', 'g4'), 365, replace = TRUE),
  date = sample(seq(as.Date('2023-01-01'), as.Date('2023-12-31'), by = "day"), 365),
  values = rnorm(365, 10, 1)
)

df2 <- data.frame(
  group = sample(c('g1', 'g2', 'g3', 'g4'), 365, replace = TRUE),
  date = sample(seq(as.Date('2023-01-01'), as.Date('2023-12-31'), by = "day"), 365),
  values = rnorm(365, 10, 1)
)

这会创建两个随机数据框，每个日期都有一个随机值，但仅针对一组，因此缺少日期。下面，我使用 data.table 函数将两个数据帧组合在一起，填充缺失的日期，并使用

runner

 包计算每组的平均运行量。但我注意到，值为 0 的日期不计入平均值。例如，此代码生成的数据框：

df <- rbindlist(list(df1, df2)) 

idx <- df[,.(date = seq(min(date), max(date), "day")), by = group]
setkey(df, group, date)
setkey(idx, group, date)

df <- df[idx] %>%
  setorder(date) %>%
  .[,.(
    date = lubridate::ymd(date),
    values = ifelse(is.na(values), 0, as.numeric(values)),
    running_values = runner::mean_run(x = values, k = 90, idx = date)
  ), by = group]

产生如下所示的数据框：

组日期价值观运行值g12023-01-049.4129.412g12023-01-0509.412g12023-01-0609.412g12023-01-0709.412g12023-01-0810.78810.100g12023-01-09010.100

但是，我期望的输出（使用 tidyverse 函数代替）是：

df <- bind_rows(df1, df2) %>%
  complete(
    nesting(group),
    date = seq.Date(min(date), max(date), by = "day")
  ) %>%
  group_by(group) %>%
  arrange(date) %>%
  mutate(
    values = ifelse(is.na(values), 0, as.numeric(values)),
    date = lubridate::ymd(date),
    running_values = runner::mean_run(x = values, k = 90, idx = date)
  )

这给出了我想要的数据框。

组日期价值观运行值g12023-01-0100g12023-01-0200g12023-01-0300g12023-01-049.4122.353g12023-01-0501.882g12023-01-0601.568

使用 tidyverse 方法还会使组在所有组中最早的日期开始，而 data.table 将仅使用每个组

每个组的最早日期，尽管这不是我主要关心的。如何使用 data.table 函数重现 tidyverse 表，这样我就不必加载这些依赖项？一个简单的解决方案似乎是将 NA 转换为任意小的数字 (0.000000001)，这似乎工作得很好，但我想了解为什么会出现这种差异。

df1 <- data.frame(
  group = sample(c('g1', 'g2', 'g3', 'g4'), 365, replace = TRUE),
  date = sample(seq(as.Date('2023-01-01'), as.Date('2023-12-31'), by = "day"), 365),
  values = rnorm(365, 10, 1)
)

df2 <- data.frame(
  group = sample(c('g1', 'g2', 'g3', 'g4'), 365, replace = TRUE),
  date = sample(seq(as.Date('2023-01-01'), as.Date('2023-12-31'), by = "day"), 365),
  values = rnorm(365, 10, 1)
)

# Bad output (data.table method)
df <- rbindlist(list(df1, df2)) 

idx <- df[,.(date = seq(min(date), max(date), "day")), by = group]
setkey(df, group, date)
setkey(idx, group, date)

df <- df[idx] %>%
  setorder(date) %>%
  .[,.(
    date = lubridate::ymd(date),
    values = ifelse(is.na(values), 0, as.numeric(values)),
    running_values = runner::mean_run(x = values, k = 90, idx = date)
  ), by = group]

# Desired output (tidyverse method)
df <- bind_rows(df1, df2) %>%
  complete(
    nesting(group),
    date = seq.Date(min(date), max(date), by = "day")
  ) %>%
  group_by(group) %>%
  arrange(date) %>%
  mutate(
    values = ifelse(is.na(values), 0, as.numeric(values)),
    date = lubridate::ymd(date),
    running_values = runner::mean_run(x = values, k = 90, idx = date)
  )

Answer 1

CJ()

创建一个填充面板（每个组每天）并将其右连接到数据：

df <- rbind(df1,df2)[
  CJ(group=unique(df$group), date=seq(min(df$date), max(df$date), "day")),
  on=.(group, date)
  ][is.na(values), values:=0]

然后进行计算：

df[, running_values := runner::mean_run(x = values, k = 90, idx = date), by=.(group)]