Question

我有一项任务可以对时间序列数据进行异常检测。我有异常检测的代码，但我正在尝试为它准备数据。数据如下所示。

timestampUtc

2016-08-01 14:38:01, 2016-08-01 14:38:06, 2016-08-01 14:38:12, 2016-08-01 14:38:18, 2016-08-01 14:38:22, 2016-08-01 14:38:27, 2016-08-01 14:38:27, 2016-08-01 14:38:30, 2016-08-01 14:38:37, 2016-08-01 14:38:38, 2016-08-01 14:38:38, 2016-08-01 14:38:46, 2016-08-01 14:39:03, 2016-08-01 14:39:03, 2016-08-01 14:39:10, 2016-08-01 14:39:12, 2016-08-01 14:39:15, 2016-08-01 14:39:16, 2016-08-01 14:39:20, 2016-08-01 14:39:28

首先，我想在timestampUtc列中将秒数设为零。接下来，我想创建一个显示count的列，并希望计算该特定分钟的值数。例如，输出应如下所示：

timestampUtc count

2016-08-01 14:38:00 12, 2016-08-01 14:39:00 6, 2016-08-01 14:40:00 8

Answer 1

您可以使用as.POSIXct()将字符串转换为日期，使用某种格式忽略秒数，然后使用table进行汇总：

timestampUtc <- c('2016-08-01 14:38:01', '2016-08-01 14:38:06', '2016-08-01 14:38:12', '2016-08-01 14:38:18', '2016-08-01 14:38:22', '2016-08-01 14:38:27', '2016-08-01 14:38:27', '2016-08-01 14:38:30', '2016-08-01 14:38:37', '2016-08-01 14:38:38', '2016-08-01 14:38:38', '2016-08-01 14:38:46', '2016-08-01 14:39:03', '2016-08-01 14:39:03', '2016-08-01 14:39:10', '2016-08-01 14:39:12', '2016-08-01 14:39:15', '2016-08-01 14:39:16', '2016-08-01 14:39:20', '2016-08-01 14:39:28')
timestampUtc <- as.POSIXct(timestampUtc, format="%Y-%m-%d %H:%M", tz="UTC")
table(timestampUtc)
2016-08-01 14:38:00 2016-08-01 14:39:00 
                 12                   8

Answer 2

假设您的时间戳已经是POSIXt格式，并且您的时间戳数据存储在df中 -

df$count <- 1
df$timestamp <- format(df$timestamp, format = "%Y-%m-%d %H:%M")
df <- aggregate(count ~ timestamp, data = df, FUN = sum)
names(df) <- c("timestamp", "count")

Answer 3

POSIXt类的cut和seq方法都有breaks（或by）的间隔选项：

 timestampUtc <-scan(text="2016-08-01 14:38:01, 2016-08-01 14:38:06, 2016-08-01 14:38:12, 2016-08-01 14:38:18, 2016-08-01 14:38:22, 2016-08-01 14:38:27, 2016-08-01 14:38:27, 2016-08-01 14:38:30, 2016-08-01 14:38:37, 2016-08-01 14:38:38, 2016-08-01 14:38:38, 2016-08-01 14:38:46, 2016-08-01 14:39:03, 2016-08-01 14:39:03, 2016-08-01 14:39:10, 2016-08-01 14:39:12, 2016-08-01 14:39:15, 2016-08-01 14:39:16, 2016-08-01 14:39:20, 2016-08-01 14:39:28",
                      what="", sep=",")
#Read 20 items

table( cut( as.POSIXct(timestampUtc), breaks="min")  )
#------------
2016-08-01 14:38:00 2016-08-01 14:39:00 
                 12                   8

如果你想要10或15分钟的间隔，它可能已经过了＃10; 10分钟＆＃34;或＆＃34; 15分钟＆＃34;。到目前为止，其他一个答案中删除了输入阶段的信息，我认为这是一个值得怀疑的做法，但code_is_entropy在传递给format的阶段使用了table缩短的格式字符串。

准备数据以进行异常检测

3 个答案: