R - 将事件日志（异步日志）转换为时间序列（同步日志）

Posted 2021-04-11

好奇，如果你们都有一个更有效或更优雅的方式将事件日志转换为时间序列。

并不是那么整齐的中心，但好奇你是否有一个漂亮的整齐方法的头脑？我试图利用dplyr :: mutate的滞后函数在值为NA时进行前瞻性观察，但我似乎无法重复滞后。

这是一个简单的例子

library(dplyr)
set.seed(1)
events <- tibble(
  t = runif(10, 0, 100) %>% sort(),
  value = runif(10, 0, 1)
)

events
# A tibble: 10 x 2
           t     value
       <dbl>     <dbl>
 1  6.178627 0.2059746
 2 20.168193 0.1765568
 3 26.550866 0.6870228
 4 37.212390 0.3841037
 5 57.285336 0.7698414
 6 62.911404 0.4976992
 7 66.079779 0.7176185
 8 89.838968 0.9919061
 9 90.820779 0.3800352
10 94.467527 0.7774452

这是一个超级hacky方式来做到这一点。时间序列事件：

accordian <- function(events_data, freq = 1){
  t_seq = seq(
    from = min(events_data$t)-freq %>% round(0), 
    to = max(events_data$t) + freq, 
    by = freq)
  timeseries = tibble(
    t = t_seq,
    value = NA
  )
  timeseries = bind_rows(
    events_data,
    timeseries
  ) %>%
    arrange(t) 
  for (i in 2:length(timeseries$value)){
    if (is.na(timeseries$value[i])){ timeseries$value[i] = timeseries$value[i-1] }
  }
  timeseries = timeseries %>%
    filter(t %in% t_seq)

  return(timeseries)
}

accordian(events)
# A tibble: 102 x 3
           t     value       type
       <dbl>     <dbl>      <chr>
 1  6.178627 0.2059746 events log
 2 20.168193 0.1765568 events log
 3 26.550866 0.6870228 events log
 4 37.212390 0.3841037 events log
 5 57.285336 0.7698414 events log
 6 62.911404 0.4976992 events log
 7 66.079779 0.7176185 events log
 8 89.838968 0.9919061 events log
 9 90.820779 0.3800352 events log
10 94.467527 0.7774452 events log
# ... with 92 more rows

并根据事件日志明确事件日志和时间序列之间的区别：

library(ggplot2)
bind_rows(
  events %>%
    mutate(type = "events log"),
  accordian(events) %>%
    mutate(type = "time series"),
) %>%
ggplot(
  aes(x = t, y = value, color = type)
) +
  geom_line()

我很想得到你的建议！

答案

这不是你问题的答案，但根据你的评论，我想向你展示我的基础R方法。也许我们可以把它翻译成一个整齐的解决方案：

async_to_sync <- function(async_df) {
  # Creates a synchronous (multivariate) time series from an asynchronous event log.
  # It is assumed the asynchronous log contains no missing values.

  # creates simple time sequence, could be modified to have a different freuency
  time_sequence <- seq(min(async_df$time), max(async_df$time), by = 1)

  # constructing a new empty dataframe:
  sync_df <- data.frame(matrix(ncol = ncol(async_df), nrow = length(time_sequence)))
  colnames(sync_df) <- colnames(async_df)
  sync_df$time <- time_sequence

  # Fill in already known values in the synchronous time series:
  for(i in 1:nrow(async_df)) {
    time_value <- async_df$time[i]
    sync_df[which(sync_df$time == time_value, arr.ind = TRUE), ] <- async_df[i, ] 
  }

  # Filling in the blanks:
  for(i in 1:nrow(sync_df)) {
    if(sync_df[i, ] %>% is.na %>% any) {
      sync_df[i, ] <- sync_df[i - 1, ]
    }
  }

  return(sync_df)
}

在一个例子上运行这个：

> set.seed(1234)
> async_df <- data.frame(time = c(2, 6, 7, 14), x1 = LETTERS[1:4], x2 = 1:4, stringsAsFactors = FALSE)
> async_df
  time x1 x2
1    2  A  1
2    6  B  2
3    7  C  3
4   14  D  4
> async_to_sync(async_df = async_df)
   time x1 x2
1     2  A  1
2     2  A  1
3     2  A  1
4     2  A  1
5     6  B  2
6     7  C  3
7     7  C  3
8     7  C  3
9     7  C  3
10    7  C  3
11    7  C  3
12    7  C  3
13   14  D  4