生成用于多个ACF的箱线图 [英] Produce a boxplot for multiple ACFs

问题描述

我使用以下命令在大约200列上运行 forecast :: Acf 。现在，我想生成一个箱形图，显示滞后1:36时相关值的分布。

I used the following to run forecast::Acf over about 200 columns. Now I would like to generate a boxplot showing the distribution of correlation values at lag 1:36.

## a simple example
d <- data.frame(ts1 = rnorm(100), ts2 = rnorm(100))
acorr <- apply(d, 2, Acf)

我现在想要的是一个箱形图，其中x值为1,2，y值为ACF， ts1 和 ts2 。

What I now want is a boxplot where x-values are 1,2 and the y-values are ACF for ts1 and ts2.

推荐答案

假设您有多个时间序列存储在数据帧 d 中（每一列是一个序列），我们可以使用以下方法获得直到滞后36的ACF（ nrow （d）>> 36 才有意义！）：

Suppose you have multiple time series stored in a data frame d (each column is one series), we can use the following to obtain ACF up to lag 36 (nrow(d) >> 36 to make sense!):

## for data frame `d`
acfs <- sapply(d, function (u) c(acf(u, lag.max = 36, plot = FALSE)$acf)[-1])

• R基本函数 acf 工作;设置 lag.max = 36 和 plot = FALSE ;


• acf 返回一个列表，我们想要 $ acf 字段。请注意，这是一个3D光栅，因此我们想使用 c（）;


• ACF滞后0将其展平为向量是1并且不感兴趣，因此我们将其删除 [-1] ;


• sapply 将返回一个矩阵，每一列为每个序列提供ACF。


• R base function acf is sufficient for the job; set lag.max = 36 and plot = FALSE;
• acf returns a list, and we want $acf field. Note, this is a 3D arrary, so we want to flatten it into a vector using c();
• ACF at lag 0 is 1 and is not of interest, so we drop it by [-1];
• sapply would return a matrix, each column giving ACF for each series.

如果您将时间序列存储在矩阵（普通矩阵或带有 mts类的矩阵），我们使用 apply 而不是 sapply ：

In case you have time series stored in a matrix (either an ordinary matrix or one with "mts" class), we use apply rather than sapply:

## for matrix `d`
acfs <- apply(d, 2L, function (u) c(acf(u, lag.max = 36, plot = FALSE)$acf)[-1])

要生成箱线图，只需使用：

To produce a boxplot, simply use:

boxplot(acfs)

为什么 $ acf 是3D数组。因为 acf 函数可以直接处理多个时间序列。尝试：

Why is $acf a 3D array. Because acf function can handle multiple time series directly. Try:

## whether `d` is data frame or matrix, it is converted to "mts" inside `acf`
oo <- acf(d, lag.max = 36, plot = FALSE)$acf

在这种情况下，问题在于还要计算互相关（CCF）。

The problem is, in this case, cross-correlation (CCF) is also computed.




在x轴上，我想要1-36，而不是 ts1 和 ts2 。我需要时间序列上每个滞后的分布。如果可以解决，您的答案将是非常好的。

On the x-axis I want 1-36, not ts1 and ts2. I need the distribution of at each lag over time series. If you can fix that your answer is very good.

Ei？我读错了你的问题吗？好吧，在那种情况下，您只需 boxplot acfs 的转置：

Ei? Did I misread your question. Well, in that case, you just boxplot the transpose of acfs:

boxplot(t(acfs))

这篇关于生成用于多个ACF的箱线图的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！