我一直在尝试将滚动均值应用于数据框中的多个列，其中每列包含来自多个人的数据。我已成功使用RcppRoll包中的roll_mean并使用lapply。我在下面的例子中包含了一个使用虚拟数据帧和输出的例子。
x <- rnorm(20,1);
y <- rnorm(20,2);
z <- rnorm(20,3);
ID <- rep(1:2, each=10);
mydf <- data.frame(ID, x, y, z);
vars <- c("x", "y", "z");
setDT(mydf)[, paste0(vars, "_", "mean") := lapply(.SD, function(x) roll_mean(x, n=3, na.rm = TRUE)), .SDcols = vars, by = ID]
mydf
ID x y z x_mean y_mean z_mean
1: 1 0.34457704 1.9580361 2.6458335 1.2515642 1.8307447 2.569645
2: 1 1.41839352 2.0697324 1.8495358 1.7012511 1.7248261 2.988908
3: 1 1.99172192 1.4644657 3.2135652 1.8455087 1.7165419 3.184736
4: 1 1.69363783 1.6402801 3.9036227 1.5002658 2.1512764 3.289555
5: 1 1.85116646 2.0448798 2.4370206 0.9775842 3.1215589 2.563110
6: 1 0.95599300 2.7686692 3.5280206 0.8477701 3.4576141 3.106095
7: 1 0.12559300 4.5511275 1.7242892 0.9450234 3.5134499 3.020176
8: 1 1.46172438 3.0530454 4.0659766 0.9080677 3.0100022 3.371839
9: 1 1.24775283 2.9361768 3.2702614 1.2515642 1.8307447 2.569645
10: 1 0.01472603 3.0407845 2.7792776 1.7012511 1.7248261 2.988908
11: 2 -0.91146047 2.5898074 2.0328348 0.4314443 1.2688530 2.477879
12: 2 0.48183559 1.8230335 2.6910075 1.2689767 0.9650435 2.544006
13: 2 1.72395769 -0.6062819 2.7097949 0.8747931 1.2273766 1.974265
14: 2 1.60113680 1.6783790 2.2312143 0.2579207 1.6945497 2.233321
15: 2 -0.70071522 2.6100328 0.9817857 0.1162224 2.0928536 2.606608
16: 2 -0.12665946 0.7952374 3.4869635 1.3884888 2.1063817 2.986786
17: 2 1.17604187 2.8732906 3.3510742 2.0557599 2.2701173 3.178248
18: 2 3.11608400 2.6506171 2.1223190 1.5553274 2.3987061 3.015501
19: 2 1.87515393 1.2864441 4.0613513 0.4314443 1.2688530 2.477879
20: 2 -0.32525560 3.2590570 2.8628313 1.2689767 0.9650435 2.544006
从输出表（mydf）可以看出，平均参数已经作为lapply语句的一部分创建，并且已经为每个单独的ID计算了滚动平均值。但是，滚动平均函数已循环结果以填充数据框，因为roll_mean函数从每个单独ID的10个原始值生成8个值。它使用回收来填充每个ID的最后两行。 我的实际数据是时间序列数据，我不希望结果被回收。我希望通过将原始x值添加到x_mean列的开头直到有足够的原始数据来产生3点滚动平均值来避免回收。
我已尝试搜索（在SO和Google上）有关避免在roll_mean或类似功能中回收的帖子，但没有成功。
有没有人知道如何在我的示例中填充前两行以避免在roll_mean函数中进行回收？
感谢。
答案 0 :(得分：0)
整个解决方案：
x <- rnorm(20,1);
y <- rnorm(20,2);
z <- rnorm(20,3);
ID <- rep(1:2, each=10);
mydf <- data.table(ID, x, y, z); # Changed to dt here
vars <- c("x", "y", "z");
# fill = NA and align = 'right'
mydf[, paste0(vars, "_", "mean") := lapply(.SD, function(x) RcppRoll::roll_mean(x, n = 3, na.rm = TRUE, fill = NA, align = 'right')), .SDcols = vars, by = ID]
mydf
# ID x y z x_mean y_mean z_mean
# 1: 1 0.3735462 2.9189774 2.835476 NA NA NA
# 2: 1 1.1836433 2.7821363 2.746638 NA NA NA
# 3: 1 0.1643714 2.0745650 3.696963 0.5738536 2.591893 3.093026
# 4: 1 2.5952808 0.0106483 3.556663 1.3144318 1.622450 3.333422
# 5: 1 1.3295078 2.6198257 2.311244 1.3630533 1.568346 3.188290
# ...
mydf[is.na(x_mean), c(paste0(vars, "_", "mean")) := mget(paste0(vars))]
mydf
# ID x y z x_mean y_mean z_mean
# 1: 1 0.3735462 2.9189774 2.835476 0.3735462 2.918977 2.835476
# 2: 1 1.1836433 2.7821363 2.746638 1.1836433 2.782136 2.746638
# 3: 1 0.1643714 2.0745650 3.696963 0.5738536 2.591893 3.093026
# 4: 1 2.5952808 0.0106483 3.556663 1.3144318 1.622450 3.333422
# 5: 1 1.3295078 2.6198257 2.311244 1.3630533 1.568346 3.188290
# ...
修改
mydf的遗漏部分也可以填充一点“更聪明”＃34;方式，即在每次迭代中使用滚动装置，窗口小1：
for (n_inner in n_roll:1) {
mydf[!complete.cases(mydf),
paste0(vars, "_", "mean") := lapply(
.SD, function(x) RcppRoll::roll_mean(x, n = n_inner, na.rm = TRUE, fill = NA, align = 'right')), .SDcols = vars, by = ID]
}
# ID x y z x_mean y_mean z_mean
# 1: 1 0.3735462 2.9189774 2.835476 0.3735462 2.918977 2.835476 <- Values from x, y and z
# 2: 1 1.1836433 2.7821363 2.746638 0.7785948 2.850557 2.791057 <- roll_mean with window 2
# 3: 1 0.1643714 2.0745650 3.696963 0.5738536 2.591893 3.093026 <- roll_mean with window 3
# 4: 1 2.5952808 0.0106483 3.556663 1.3144318 1.622450 3.333422 <- as above
# 5: 1 1.3295078 2.6198257 2.311244 1.3630533 1.568346 3.188290
# ...