如何在R中绘制平滑线或95％的逐点间隔？

I have the following image: 如何在R中绘制平滑线或95％的逐点间隔？

我有以下图片:

And I would like to smooth the red and blue line. But I have no idea how to do it. The red and blue lines respectively represent upper & lower 95% intervals of the black dots. (Notice that I didnt use any regression formula to obtain the 95% intervals) I read about the loess function but when i tried to use it. I get back the same plot. So is there any particular built in R function that will allow me to smooth these 2 lines.

我想平滑红线和蓝线。但我不知道该怎么做。红线和蓝线分别代表黑点的上下95%间隔。 (请注意,我没有使用任何回归公式来获得95%的间隔)我读到了黄土函数,但是当我尝试使用它时。我得到了同样的情节。那么是否有任何特殊的内置R功能,可以让我平滑这两行。

Alternatively, is there a way to obtain a "95% point wise intervals" for this problem ?

或者,有没有办法获得这个问题的“95%点间隔”?

The code is given below:

代码如下:

residual.plot <- function(a,b)
{
    log.y1 <- log(a) - b * log(energy)
    fitted.y <- exp(log.y1)
    diff <- count - fitted.y
    #normal approximation
    low.interval <- c()
    high.interval <- c()
    for(i in 1:350)
    {
        low <- diff[i] - sqrt(  exp(log(a) - b * log(energy[i])) )*qnorm(0.975)
        high <- diff[i] + sqrt(  exp(log(a) - b * log(energy[i]))  )*qnorm(0.975)
        low.interval <- append(low.interval, low)
        high.interval <- append(high.interval, high)
    }
    par(mfrow = c(1,1))
    plot(energy, diff, ylim = c(-10,10), type = "p", pch = 7)
    lines(energy, low.interval, type = "p", col = "red", pch = 1)
    lines(energy, high.interval, type = "p", col = "blue", pch = 1)


}

1 个解决方案

#1

First of all, never ever dare posting code like that again. You commit two mortal sins :

首先,再也不敢发布这样的代码了。你犯了两个致命的罪:

you grow objects in an iterative loop (tons of problems there)

你在一个迭代循环中增长对象(那里有很多问题)

you don't use the fact that R works vectorized.

你没有使用R工作矢量化的事实。

This said, the easiest way of doing this is by using lowess, provided there's no NA values in your data. Your function should be then something like this :

这就是说,最简单的方法是使用lowess,前提是数据中没有NA值。你的功能应该是这样的:

residual.plot <- function(a,b,count,energy)
{
    log.y1 <- log(a) - b * log(energy)
    fitted.y <- exp(log.y1)
    diff <- count - fitted.y

    #normal approximation
        low <- diff - sqrt(  exp(log(a) - b * log(energy)) )*qnorm(0.975)
        high <- diff + sqrt(  exp(log(a) - b * log(energy))  )*qnorm(0.975)

    par(mfrow = c(1,1))
    plot(energy, diff, ylim = c(-10,10), type = "p", pch = 7)
    lines(lowess(energy, low), type = "p", col = "red", pch = 1)
    lines(lowess(energy, high), type = "p", col = "blue", pch = 1)


}

PS: To make a function useful, you shouldn't count on variables from outside the function like for example count and energy. Add them as an argument to the function, so you can use the function later on when using a different dataset.

PS:为了使函数有用,你不应该依赖函数外部的变量,例如count和energy。将它们作为参数添加到函数中,以便稍后在使用其他数据集时可以使用该函数。

#1