r - 一列中有多个日期：R

Question

好的，我已经解决了几个问题，我认为我在数据中看到了一个可以解决的模式（也就是说，我看到我可以区分 dmy 和 mdy）。

让我介绍一下数据集的背景。该数据集包含当前隔离期间旅行的人员的信息。因此，当我向下滚动数据集时，我看到大多数日期以 MDY 格式输入，而有些则以 dmy 格式输入。

所以对于这个数据集，（数据集包含直到八月的信息）。歧义可以通过以下方式解决：

1. 日期不能是未来。因此，这解决了 08-12-2020 和 12-08-2020 之间的差异。另外，我只有到六月的数据。
2. 日子比几个月快。如果我看到一个序列（无论它是 dmy 还是 mdy），如果我看到一个数字每隔几行发生变化（比如说 20），那么我知道正在变化的数字是一天而不是一个月。例子：

在这种情况下，我将如何正确分配日期？

标签： rdplyrlubridate

Answer 1

因此，如果我理解正确，您的数据是按时间顺序排列的，并且您对测量数据的时间范围有所了解。

鉴于此，这是我创建加扰数据然后重新解析它的尝试：

library(dplyr)
library(lubridate)
library(zoo) # for na.locf function

# create some scrambled data to work with
df <- tibble(
    date_ground_truth = rep(seq(from = ymd('20190801'), to = ymd('20200730'), by = 1), each = 5)
  ) %>%
  mutate(
    date_inconsistent_chr = format(date_ground_truth, ifelse(runif(nrow(.), 0, 1) > 0.3, '%Y/%m/%d', '%Y/%d/%m'))
  )

# providing here the date range in which your observations lie. I only know that it maxes end of July 2020, so my end result has some remaining unknowns at the start
daterange_known_min <- NA_Date_
daterange_known_max <- ymd('20200730')

# initiate a cleaned df - for any date where we have a day > 12, we know that it can only be one format of YMD/YDM
df_recleaned <- df %>%
  mutate(
    date_parsed_ymd = as.Date(date_inconsistent_chr, '%Y/%m/%d'), # try YMD
    date_parsed_ydm = as.Date(date_inconsistent_chr, '%Y/%d/%m'), # try YDM
    
    date_parsed_deducted = case_when( # write out the clear cut cases
      day(date_parsed_ymd) > 12 ~ date_parsed_ymd,
      day(date_parsed_ydm) > 12 ~ date_parsed_ydm,
      date_parsed_ymd == date_parsed_ydm ~ date_parsed_ymd,
      T ~ NA_Date_
    )
  )

# we will run over the data until we can not deduct any more new dates from what we've learnt so far:

new_guesses_possible <- T
while(new_guesses_possible) {
  # how many dates did we already deduct?
  num_deducted_dates_before <- df_recleaned %>% filter(!is.na(date_parsed_deducted)) %>% nrow()
  
  # deduct more, based on our knowledge that the dates are chronological and within a certain time-frame
  df_recleaned <- df_recleaned %>%
    mutate(
      earliest_possible_date = coalesce(na.locf(date_parsed_deducted, na.rm = F), daterange_known_min),
      last_possible_date = coalesce(na.locf(date_parsed_deducted, fromLast = T, na.rm = F), daterange_known_max),
      
      ymd_guess_in_range = coalesce(date_parsed_ymd >= earliest_possible_date & date_parsed_ymd <= last_possible_date, F),
      ydm_guess_in_range = coalesce(date_parsed_ydm >= earliest_possible_date & date_parsed_ydm <= last_possible_date, F),
      
      date_parsed_deducted = case_when(
        # keep clear cases
        !is.na(date_parsed_deducted) ~ date_parsed_deducted,
        
        # if the ymd-guess falls between the last clear case and next clear case, take ymd
        ymd_guess_in_range & !ydm_guess_in_range ~ date_parsed_ymd,
        # same approach for ydm
        ydm_guess_in_range & !ymd_guess_in_range ~ date_parsed_ydm,
        
        # cover the cases where we don't know either the last or next clear parsed date
        # if one of the parsed dates falls before the "first possible date", take the other one.
        #   (if no daterange_known_min is given, these rows will result in NA and not do anything...)
        date_parsed_ymd < daterange_known_min ~ date_parsed_ydm,
        date_parsed_ydm < daterange_known_min ~ date_parsed_ymd,
        # inversely, if one parsed option falls after the "last possible date", ignore it.
        date_parsed_ymd > daterange_known_max ~ date_parsed_ydm,
        date_parsed_ydm > daterange_known_max ~ date_parsed_ymd
      )
    )
  
  # how many dates did we now deduct?
  num_deducted_dates_after <- df_recleaned %>% filter(!is.na(date_parsed_deducted)) %>% nrow()
  
  # do we need to go on?
  new_guesses_possible <- num_deducted_dates_after > 0 & num_deducted_dates_before != num_deducted_dates_after
}

# kick out all those extra columns :)
df_recleaned_final <- df_recleaned %>%
  select(
    -date_parsed_ymd, -date_parsed_ydm,
    -earliest_possible_date, -last_possible_date,
    -ymd_guess_in_range, -ydm_guess_in_range
  )

在我的示例中，这会修复 2019 年 8 月第一周之后的所有日期
。如果您的数据中存在更长的空白，它可能会给您不同的结果。