R:如何从ggplot2中更顺畅地删除异常值?

时间:2010-04-10 06:27:03

标签: r ggplot2 statistics outliers

我有以下数据集,我试图用ggplot2绘图,它是三个实验A1,B1和C1的时间序列,每个实验有三个重复。

我正在尝试添加一个stat,它会在返回更平滑(均值和方差?)之前检测并删除异常值。我已经编写了自己的异常函数(未显示),但我希望已经有一个函数来执行此操作,我只是没有找到它。

我从ggplot2书中的一些例子看过stat_sum_df(“median_hilow”,geom =“smooth”),但是我不明白Hmisc的帮助文档,看它是否删除了异常值。

是否有一个函数可以在ggplot中删除这样的异常值,或者在下面修改我的代码以添加我自己的函数?

编辑:我刚看到这个(How to use Outlier Tests in R Code)并注意到Hadley建议使用强大的方法,例如rlm。我正在绘制细菌生长曲线,所以我不认为线性模型是最好的,但是在这种情况下对其他模型或使用或使用稳健模型的任何建议都将受到赞赏。

library (ggplot2)  

data = data.frame (day = c(1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7,1,3,5,7), od = 
c(
0.1,1.0,0.5,0.7
,0.13,0.33,0.54,0.76
,0.1,0.35,0.54,0.73
,1.3,1.5,1.75,1.7
,1.3,1.3,1.0,1.6
,1.7,1.6,1.75,1.7
,2.1,2.3,2.5,2.7
,2.5,2.6,2.6,2.8
,2.3,2.5,2.8,3.8), 
series_id = c(
"A1", "A1", "A1","A1",
"A1", "A1", "A1","A1",
"A1", "A1", "A1","A1",
"B1", "B1","B1", "B1",
"B1", "B1","B1", "B1",
"B1", "B1","B1", "B1",
"C1","C1", "C1", "C1",
"C1","C1", "C1", "C1",
"C1","C1", "C1", "C1"),
replicate = c(
"A1.1","A1.1","A1.1","A1.1",
"A1.2","A1.2","A1.2","A1.2",
"A1.3","A1.3","A1.3","A1.3",
"B1.1","B1.1","B1.1","B1.1",
"B1.2","B1.2","B1.2","B1.2",
"B1.3","B1.3","B1.3","B1.3",
"C1.1","C1.1","C1.1","C1.1",
"C1.2","C1.2","C1.2","C1.2",
"C1.3","C1.3","C1.3","C1.3"))

> data
   day   od series_id replicate
1    1 0.10        A1      A1.1
2    3 1.00        A1      A1.1
3    5 0.50        A1      A1.1
4    7 0.70        A1      A1.1
5    1 0.13        A1      A1.2
6    3 0.33        A1      A1.2
7    5 0.54        A1      A1.2
8    7 0.76        A1      A1.2
9    1 0.10        A1      A1.3
10   3 0.35        A1      A1.3
11   5 0.54        A1      A1.3
12   7 0.73        A1      A1.3
13   1 1.30        B1      B1.1
... etc...

这是我到目前为止所做的工作,并且工作得很好,但不会删除异常值:

r <- ggplot(data = data, aes(x = day, y = od))
r + geom_point(aes(group = replicate, color = series_id)) + # add points
   geom_line(aes(group = replicate, color = series_id)) + # add lines
   geom_smooth(aes(group = series_id))  # add smoother, average of each replicate

编辑:我刚刚在下面添加了两个图表,显示了我从实际数据而不是上面的示例数据中获得的离群值问题的示例。

第一个图显示了系列p26s4,在第32天左右,两个重复中出现了一些非常奇怪的事情,显示出2个异常值。

第二个图显示了系列p22s5,在第18天,当天读数有些奇怪,我认为可能是机器错误。

目前,我正在关注数据,检查增长曲线是否正常。在采纳哈德利的建议并设定家庭=“对称”之后,我相信黄土更顺畅,可以忽略异常值。

p26s4 shows around day 32 something really weird went on in two of the replicates, showing 2 outliers http://img696.imageshack.us/img696/8743/p26s4loess.png p22s5 shows that on day 18, something weird went on with the reading that day, likely machine error I think http://img521.imageshack.us/img521/8083/p22s5loess.png

@Peter / @ hadley,接下来我要做的是尝试将逻辑,gompertz或richard的增长曲线拟合到这个数据而不是黄土,并计算指数阶段的增长率。最终我打算在R(http://cran.r-project.org/web/packages/grofit/index.html)中使用grofit包,但是现在我想使用ggplot2手动绘制这些包。如果您有任何指示,那将非常感激。

2 个答案:

答案 0 :(得分:14)

您是否尝试过family = "symmetric" geom_smooth参数{(1}}(后者又传递给loess)?这将使黄土光滑抵抗异常值。

但是,查看您的数据,为什么您认为线性拟合不充分?你只有4个x值,而且似乎没有强有力的证据表明偏离线性。

答案 1 :(得分:2)

首先,我不确定在这些小数据上是否正确定义了“异常值”。

其次,你必须决定你所说的“异常值”是什么意思,它是药物之一,重复之一,还是其中一个时间点?

正如哈德利指出的那样,几乎没有证据表明偏离线性。

最后,我认为使用更平滑的一点是,如果有足够的数据,它可以很好地处理异常值。但你很少。

所以,我必须问你为什么要删除异常值。也就是说,你打算用这些数据做什么(除了制作好的情节)?

我希望这会有所帮助