R 如何使用data.table计算日内数据每天的滚动分位数

我想用数据表计算一个滚动分位数，它包含了几个组的数据，对于每个组，我有多天的时间，在每一天内我有多个观察。我不想计算表中每一个观察值的滚动分位数，但总是获取最后，比如说3天的数据，计算一个分位数，然后继续

我有这样的数据：

test2 <- data.table(group = rep(c("a", "b"), each = 100),
                    date = rep(rep(seq(from = as.Date('2017-01-01'),
                                    as.Date('2017-01-10'),
                                    by = "day"), each = 10), 2),
                    time = rep(rep(seq(from = 1, 10, by = 1), times = 10), 2),
                    some_data = rnorm(200) + c(1:20, 20:1, 30:1, 1:30, 30:1, 1:20, 20:1, 1:30))

tests_result <- test2[, list(date = date,
                         q_30 = rollapply(some_data,
                                          30, quantile,
                                          probs = 0.3,
                                          fill = NA, align = "right")),
                  by = "group"][seq(from = 10, to = 200, by = 10)]

    group       date      q_30
 1:     a 2017-01-01        NA
 2:     a 2017-01-02        NA
 3:     a 2017-01-03 10.284081
 4:     a 2017-01-04  8.281827
 5:     a 2017-01-05  8.281827
 6:     a 2017-01-06  8.281827
 7:     a 2017-01-07 10.274793
 8:     a 2017-01-08  4.749455
 9:     a 2017-01-09  4.749455
10:     a 2017-01-10  9.246267
11:     b 2017-01-01        NA
12:     b 2017-01-02        NA
13:     b 2017-01-03 10.145996
14:     b 2017-01-04  5.423782
15:     b 2017-01-05  5.423782
16:     b 2017-01-06  9.741683
17:     b 2017-01-07 10.123940
18:     b 2017-01-08  4.347293
19:     b 2017-01-09  4.347293
20:     b 2017-01-10  9.177718

总结挑战：

test2 <- data.table(group = rep(c("a", "b"), each = 100),
                    date = rep(rep(seq(from = as.Date('2017-01-01'),
                                    as.Date('2017-01-10'),
                                    by = "day"), each = 10), 2),
                    time = rep(rep(seq(from = 1, 10, by = 1), times = 10), 2),
                    some_data = rnorm(200) + c(1:20, 20:1, 30:1, 1:30, 30:1, 1:20, 20:1, 1:30))

tests_result <- test2[, list(date = date,
                         q_30 = rollapply(some_data,
                                          30, quantile,
                                          probs = 0.3,
                                          fill = NA, align = "right")),
                  by = "group"][seq(from = 10, to = 200, by = 10)]

    group       date      q_30
 1:     a 2017-01-01        NA
 2:     a 2017-01-02        NA
 3:     a 2017-01-03 10.284081
 4:     a 2017-01-04  8.281827
 5:     a 2017-01-05  8.281827
 6:     a 2017-01-06  8.281827
 7:     a 2017-01-07 10.274793
 8:     a 2017-01-08  4.749455
 9:     a 2017-01-09  4.749455
10:     a 2017-01-10  9.246267
11:     b 2017-01-01        NA
12:     b 2017-01-02        NA
13:     b 2017-01-03 10.145996
14:     b 2017-01-04  5.423782
15:     b 2017-01-05  5.423782
16:     b 2017-01-06  9.741683
17:     b 2017-01-07 10.123940
18:     b 2017-01-08  4.347293
19:     b 2017-01-09  4.347293
20:     b 2017-01-10  9.177718

每天只计算一次分位数，而不是10分位数 时代

在给定天数内执行分位数计算 即使每天观察的次数不同。 i、 如果我想计算分位数，基于2天，第一天 将有10个值，第二天20个值，我会得到一个基于结果的结果 这两天的30个数值以及最终结果 计算将分配到第二天的日期

编辑：

test2 <- data.table(group = rep(c("a", "b"), each = 100),
                    date = rep(rep(seq(from = as.Date('2017-01-01'),
                                    as.Date('2017-01-10'),
                                    by = "day"), each = 10), 2),
                    time = rep(rep(seq(from = 1, 10, by = 1), times = 10), 2),
                    some_data = rnorm(200) + c(1:20, 20:1, 30:1, 1:30, 30:1, 1:20, 20:1, 1:30))

tests_result <- test2[, list(date = date,
                         q_30 = rollapply(some_data,
                                          30, quantile,
                                          probs = 0.3,
                                          fill = NA, align = "right")),
                  by = "group"][seq(from = 10, to = 200, by = 10)]

    group       date      q_30
 1:     a 2017-01-01        NA
 2:     a 2017-01-02        NA
 3:     a 2017-01-03 10.284081
 4:     a 2017-01-04  8.281827
 5:     a 2017-01-05  8.281827
 6:     a 2017-01-06  8.281827
 7:     a 2017-01-07 10.274793
 8:     a 2017-01-08  4.749455
 9:     a 2017-01-09  4.749455
10:     a 2017-01-10  9.246267
11:     b 2017-01-01        NA
12:     b 2017-01-02        NA
13:     b 2017-01-03 10.145996
14:     b 2017-01-04  5.423782
15:     b 2017-01-05  5.423782
16:     b 2017-01-06  9.741683
17:     b 2017-01-07 10.123940
18:     b 2017-01-08  4.347293
19:     b 2017-01-09  4.347293
20:     b 2017-01-10  9.177718

我想出了一种方法来处理数据集的大小。但是我认为它仍然可以改进，所以，如果你有任何建议，我想听听

我对样本数据集的处理方法如下所示：

首先计算随后每3天的观察总数，同时计算给定一天中最后一次观察的原始数据集中的行数。这些新变量将在第3行和原始行中被称为

test3 <- test2[, list(.N, orig_row = .I[.N]), by = c("group", "date")][, list(date,in_3 = rollapply(N, 3, sum, fill = NA, align = "right"),
                                                  orig_row),
                                           by = "group"]

最后，分配给聚合数据集

test3[, `:=`(q03 = quantiles)]

我也试着并行运行，但后来我的笔记本电脑的物理内存用完了，开始向磁盘写入太多内容，这比仅用一个内核运行要慢得多。
您的预期输出是什么？对于特定天数的滚动分位数是什么意思？这不是你的“理论”代码所做的。这就是为什么在代码之后我说
，然后每天做最后一次观察。我希望数据集中的每个日期都有一个数字，这将是根据当天的观察值计算出的分位数，+前几天的观察值。啊，你是对的，我提供的代码作为示例，我想要的，实际上并不完全符合我的想法。调整很快就会到来moment@mtoto我的不好，只是现在预期的结果是正确的。