根据标准从当前观察开始总结
Sum up ending with the current observation starting based on a criteria
我观察了（在下面的示例中：4）不同客户在（五）天不同的购买次数。现在我想创建一个新变量，总结每个用户在过去 20 次购买中所有用户的购买次数。
示例数据：
> da <- data.frame(customer_id = c(1,1,1,1,1,2,2,2,2,2,3,3,3,3,3,4,4,4,4,4),
+ day = c("2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15"),
+ n_purchase = c(5,2,8,0,3,2,0,3,4,0,2,4,5,1,0,2,3,5,0,3))
> da
customer_id day n_purchase
1 1 2016-04-11 5
2 1 2016-04-12 2
3 1 2016-04-13 8
4 1 2016-04-14 0
5 1 2016-04-15 3
6 2 2016-04-11 2
7 2 2016-04-12 0
8 2 2016-04-13 3
9 2 2016-04-14 4
10 2 2016-04-15 0
11 3 2016-04-11 2
12 3 2016-04-12 4
13 3 2016-04-13 5
14 3 2016-04-14 1
15 3 2016-04-15 0
16 4 2016-04-11 2
17 4 2016-04-12 3
18 4 2016-04-13 5
19 4 2016-04-14 0
20 4 2016-04-15 3
我需要知道三件事来构造我的变量：
(1) 用户一天的总购买量（日购买量）是多少？
(2) 从第一天（cumsum_day_purchases）开始，用户累计购买次数是多少？
(3) 根据当前观察，前 20 次（跨用户）购买是从哪一天开始的？这是我在编写此类变量时遇到问题的地方。
> library(dplyr)
> da %>%
+ group_by(day) %>%
+ mutate(day_purchases = sum(n_purchase)) %>%
+ group_by(customer_id) %>%
+ mutate(cumsum_day_purchases = cumsum(day_purchases))
# A tibble: 20 x 5
# Groups: customer_id [4]
customer_id day n_purchase day_purchases cumsum_day_purchases
<dbl> <fct> <dbl> <dbl> <dbl>
1 1 2016-04-11 5 11 11
2 1 2016-04-12 2 9 20
3 1 2016-04-13 8 21 41
4 1 2016-04-14 0 5 46
5 1 2016-04-15 3 6 52
6 2 2016-04-11 2 11 11
7 2 2016-04-12 0 9 20
8 2 2016-04-13 3 21 41
9 2 2016-04-14 4 5 46
10 2 2016-04-15 0 6 52
11 3 2016-04-11 2 11 11
12 3 2016-04-12 4 9 20
13 3 2016-04-13 5 21 41
14 3 2016-04-14 1 5 46
15 3 2016-04-15 0 6 52
16 4 2016-04-11 2 11 11
17 4 2016-04-12 3 9 20
18 4 2016-04-13 5 21 41
19 4 2016-04-14 0 5 46
20 4 2016-04-15 3 6 52
我现在将在以下数据集中手动计算我希望拥有的变量。
- 对于 2016-04-12 日的所有观察，我计算了累计和
通过添加购买数量来计算特定客户的购买量
当天和前一天的，因为总共
客户当天共购买了 20 件商品，
前一天。
- 对于2016-04-13这一天，我只使用某用户在
这一天，因为当天有 21 (41-20) 次新购买
导致以下输出：
> da = da %>% ungroup() %>%
+ mutate(cumsum_last_20_purchases = c(5,5+2,8,0,0+3,2,2+0,3,4,4+0,2,2+4,5,1,1+0,2,2+3,5,0,0+3))
> da
# A tibble: 20 x 6
customer_id day n_purchase day_purchases cumsum_day_purchases cumsum_last_20_purchases
<dbl> <fct> <dbl> <dbl> <dbl> <dbl>
1 1 2016-04-11 5 11 11 5
2 1 2016-04-12 2 9 20 7
3 1 2016-04-13 8 21 41 8
4 1 2016-04-14 0 5 46 0
5 1 2016-04-15 3 6 52 3
6 2 2016-04-11 2 11 11 2
7 2 2016-04-12 0 9 20 2
8 2 2016-04-13 3 21 41 3
9 2 2016-04-14 4 5 46 4
10 2 2016-04-15 0 6 52 4
11 3 2016-04-11 2 11 11 2
12 3 2016-04-12 4 9 20 6
13 3 2016-04-13 5 21 41 5
14 3 2016-04-14 1 5 46 1
15 3 2016-04-15 0 6 52 1
16 4 2016-04-11 2 11 11 2
17 4 2016-04-12 3 9 20 5
18 4 2016-04-13 5 21 41 5
19 4 2016-04-14 0 5 46 0
20 4 2016-04-15 3 6 52 3
我们可以根据 day_purchase 列超过 20 的最后一天创建一个新分组，然后在该日期上使用 cumsum：
library(dplyr)
da %>%
group_by(day) %>%
mutate(day_purchases = sum(n_purchase)) %>%
group_by(customer_id) %>%
mutate(above = with(rle(day_purchases >= 20), rep(1:length(lengths), lengths))) %>%
group_by(above, .add =TRUE) %>%
mutate(cumsum_last_20_purchases = cumsum(n_purchase))
#> # A tibble: 20 x 6
#> # Groups: customer_id, above [12]
#> customer_id day n_purchase day_purchases above cumsum_last_20_purchas…
#> <dbl> <fct> <dbl> <dbl> <int> <dbl>
#> 1 1 2016-04-11 5 11 1 5
#> 2 1 2016-04-12 2 9 1 7
#> 3 1 2016-04-13 8 21 2 8
#> 4 1 2016-04-14 0 5 3 0
#> 5 1 2016-04-15 3 6 3 3
#> 6 2 2016-04-11 2 11 1 2
#> 7 2 2016-04-12 0 9 1 2
#> 8 2 2016-04-13 3 21 2 3
#> 9 2 2016-04-14 4 5 3 4
#> 10 2 2016-04-15 0 6 3 4
#> 11 3 2016-04-11 2 11 1 2
#> 12 3 2016-04-12 4 9 1 6
#> 13 3 2016-04-13 5 21 2 5
#> 14 3 2016-04-14 1 5 3 1
#> 15 3 2016-04-15 0 6 3 1
#> 16 4 2016-04-11 2 11 1 2
#> 17 4 2016-04-12 3 9 1 5
#> 18 4 2016-04-13 5 21 2 5
#> 19 4 2016-04-14 0 5 3 0
#> 20 4 2016-04-15 3 6 3 3
由 reprex package (v0.3.0)
于 2020-07-28 创建
我观察了（在下面的示例中：4）不同客户在（五）天不同的购买次数。现在我想创建一个新变量，总结每个用户在过去 20 次购买中所有用户的购买次数。
示例数据：
> da <- data.frame(customer_id = c(1,1,1,1,1,2,2,2,2,2,3,3,3,3,3,4,4,4,4,4),
+ day = c("2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15","2016-04-11","2016-04-12","2016-04-13","2016-04-14","2016-04-15"),
+ n_purchase = c(5,2,8,0,3,2,0,3,4,0,2,4,5,1,0,2,3,5,0,3))
> da
customer_id day n_purchase
1 1 2016-04-11 5
2 1 2016-04-12 2
3 1 2016-04-13 8
4 1 2016-04-14 0
5 1 2016-04-15 3
6 2 2016-04-11 2
7 2 2016-04-12 0
8 2 2016-04-13 3
9 2 2016-04-14 4
10 2 2016-04-15 0
11 3 2016-04-11 2
12 3 2016-04-12 4
13 3 2016-04-13 5
14 3 2016-04-14 1
15 3 2016-04-15 0
16 4 2016-04-11 2
17 4 2016-04-12 3
18 4 2016-04-13 5
19 4 2016-04-14 0
20 4 2016-04-15 3
我需要知道三件事来构造我的变量： (1) 用户一天的总购买量（日购买量）是多少？ (2) 从第一天（cumsum_day_purchases）开始，用户累计购买次数是多少？ (3) 根据当前观察，前 20 次（跨用户）购买是从哪一天开始的？这是我在编写此类变量时遇到问题的地方。
> library(dplyr)
> da %>%
+ group_by(day) %>%
+ mutate(day_purchases = sum(n_purchase)) %>%
+ group_by(customer_id) %>%
+ mutate(cumsum_day_purchases = cumsum(day_purchases))
# A tibble: 20 x 5
# Groups: customer_id [4]
customer_id day n_purchase day_purchases cumsum_day_purchases
<dbl> <fct> <dbl> <dbl> <dbl>
1 1 2016-04-11 5 11 11
2 1 2016-04-12 2 9 20
3 1 2016-04-13 8 21 41
4 1 2016-04-14 0 5 46
5 1 2016-04-15 3 6 52
6 2 2016-04-11 2 11 11
7 2 2016-04-12 0 9 20
8 2 2016-04-13 3 21 41
9 2 2016-04-14 4 5 46
10 2 2016-04-15 0 6 52
11 3 2016-04-11 2 11 11
12 3 2016-04-12 4 9 20
13 3 2016-04-13 5 21 41
14 3 2016-04-14 1 5 46
15 3 2016-04-15 0 6 52
16 4 2016-04-11 2 11 11
17 4 2016-04-12 3 9 20
18 4 2016-04-13 5 21 41
19 4 2016-04-14 0 5 46
20 4 2016-04-15 3 6 52
我现在将在以下数据集中手动计算我希望拥有的变量。
- 对于 2016-04-12 日的所有观察，我计算了累计和 通过添加购买数量来计算特定客户的购买量 当天和前一天的，因为总共 客户当天共购买了 20 件商品， 前一天。
- 对于2016-04-13这一天，我只使用某用户在 这一天，因为当天有 21 (41-20) 次新购买
导致以下输出：
> da = da %>% ungroup() %>%
+ mutate(cumsum_last_20_purchases = c(5,5+2,8,0,0+3,2,2+0,3,4,4+0,2,2+4,5,1,1+0,2,2+3,5,0,0+3))
> da
# A tibble: 20 x 6
customer_id day n_purchase day_purchases cumsum_day_purchases cumsum_last_20_purchases
<dbl> <fct> <dbl> <dbl> <dbl> <dbl>
1 1 2016-04-11 5 11 11 5
2 1 2016-04-12 2 9 20 7
3 1 2016-04-13 8 21 41 8
4 1 2016-04-14 0 5 46 0
5 1 2016-04-15 3 6 52 3
6 2 2016-04-11 2 11 11 2
7 2 2016-04-12 0 9 20 2
8 2 2016-04-13 3 21 41 3
9 2 2016-04-14 4 5 46 4
10 2 2016-04-15 0 6 52 4
11 3 2016-04-11 2 11 11 2
12 3 2016-04-12 4 9 20 6
13 3 2016-04-13 5 21 41 5
14 3 2016-04-14 1 5 46 1
15 3 2016-04-15 0 6 52 1
16 4 2016-04-11 2 11 11 2
17 4 2016-04-12 3 9 20 5
18 4 2016-04-13 5 21 41 5
19 4 2016-04-14 0 5 46 0
20 4 2016-04-15 3 6 52 3
我们可以根据 day_purchase 列超过 20 的最后一天创建一个新分组，然后在该日期上使用 cumsum：
library(dplyr)
da %>%
group_by(day) %>%
mutate(day_purchases = sum(n_purchase)) %>%
group_by(customer_id) %>%
mutate(above = with(rle(day_purchases >= 20), rep(1:length(lengths), lengths))) %>%
group_by(above, .add =TRUE) %>%
mutate(cumsum_last_20_purchases = cumsum(n_purchase))
#> # A tibble: 20 x 6
#> # Groups: customer_id, above [12]
#> customer_id day n_purchase day_purchases above cumsum_last_20_purchas…
#> <dbl> <fct> <dbl> <dbl> <int> <dbl>
#> 1 1 2016-04-11 5 11 1 5
#> 2 1 2016-04-12 2 9 1 7
#> 3 1 2016-04-13 8 21 2 8
#> 4 1 2016-04-14 0 5 3 0
#> 5 1 2016-04-15 3 6 3 3
#> 6 2 2016-04-11 2 11 1 2
#> 7 2 2016-04-12 0 9 1 2
#> 8 2 2016-04-13 3 21 2 3
#> 9 2 2016-04-14 4 5 3 4
#> 10 2 2016-04-15 0 6 3 4
#> 11 3 2016-04-11 2 11 1 2
#> 12 3 2016-04-12 4 9 1 6
#> 13 3 2016-04-13 5 21 2 5
#> 14 3 2016-04-14 1 5 3 1
#> 15 3 2016-04-15 0 6 3 1
#> 16 4 2016-04-11 2 11 1 2
#> 17 4 2016-04-12 3 9 1 5
#> 18 4 2016-04-13 5 21 2 5
#> 19 4 2016-04-14 0 5 3 0
#> 20 4 2016-04-15 3 6 3 3
由 reprex package (v0.3.0)
于 2020-07-28 创建