r - 如何在 R 中的多直方图中设置统一的条形宽度

Question

我正在读取一个包含 26 列和 4288 个实例的数据集“Dummy_data.csv”，其中总共有 17 个参数（列）对我们的数据分析很重要。17 个参数中的 6 个，即（param1、param2、param3、param5、param6、param7）是关键参数，同时超出范围决定了项目是否有缺陷（类标签）。例如，

range1 = (min1, max1) = (0.25, 0.35)
range2 = (min2, max2) = (2.5, 3.1)
range3 = (min3, max3) = (680, 700)
range5 = (min5, max5) = (56, 64)
range6 = (min6, max6) = (40, 60)
range7 = (min7, max7) = (28, 38)

if (param1 out of range1 & param2 out of range2 & param3 out of range3 &
    param5 out of range5 & param6 out of range6 & param7 out of range7)
    class = 'defective'
else
    class = 'ok'

我们需要对上述数据进行两次缺陷分析。首先，我需要从项目总数中找出缺陷的百分比。其次，我需要找出 6 个关键参数中每一个的超出范围值的频率直方图，以了解这些关键参数的哪些超出范围值对缺陷品的影响更大。

我做了什么：由于这 6 个关键参数的范围大多不重叠，首先我缩放了 17 个参数（尽管缩放 6 个关键参数就足够了！）使用(x - min(x))/(max(x)-min(x))间隔，(0, 1)以便我可以进行频率分布x 轴统一刻度上 6 个参数的超出范围值。在图形方面，这意味着小于 0 的参数值意味着小于最小值，而大于 1 的参数值意味着大于最大值。因此，我从数据框中的数据集中过滤了所有有缺陷的实例，z并绘制了一个饼图来显示缺陷和正常项目的百分比。（第一次分析）

对于频率直方图（第二次分析），我将缩放数据集中的所有缺陷实例过滤scaled.dat.df到defect.dat.df. 然后我从所有 6 个参数中选择min和max来确定缺陷间隔范围。接下来，我将 6 个参数中的每一个的唯一值分箱，p1.bin.defect.dat.df并使用该函数在同一图上p7.bin.defect.dat.df绘制各个直方图。plot

多个重叠直方图的问题
我得到了如下所示的多直方图，但问题是条形的宽度因 6 个参数而异。有人知道如何为多直方图设置统一的条形宽度吗？另外，如何在这个多历史情节中添加合适的图例？

任何有用的建议/答案都将受到高度赞赏和相应的奖励。

注意：我在此处的多直方图上关注了另一个线程 how-to-plot-two-histograms-together-in-r 并想要一个与此非常相似但 6 个重叠的直方图而不是 2 个重叠的直方图（如在线程中）

library(RWeka)
library(party)
library(plyr)
library(plotrix)
library(sm)

#read data and class labels
dat <- read.csv("Dummy_data.csv", head=T, sep=",")
datm <- as.matrix(dat[,8:24])
class <- as.matrix(dat[,26])

#center and scale data
center <- c(0.25, 2.5, 680, 1067, 56, 40, 28, -99, -99, 40, 5, 50, 5000, 15000, 11.3, 9.1, 0)
scale <- c(0.1, 0.6, 20, 6, 8, 20, 10, 19, 19, 20, 2, 10, 500, 1000, 3.4, 18.3, 5)
scaled.datm <- scale(datm, center, scale)
write.table(scaled.datm, 
file = "C:\\Users\\schakrabarti\\Documents\\Dummy_data_whdr17.csv", 
append=FALSE, quote=TRUE, sep=",", eol = "\n", na = "NA", dec = ".", 
row.names = FALSE, col.names = TRUE, qmethod = c("escape", "double"),
fileEncoding = "")

#filter total non-compliants
scaled.dat.df <- as.data.frame(scaled.datm)
total <- length(scaled.dat.df[,1])
z <- c((scaled.dat.df[,"PARAM1"]<0 | scaled.dat.df[,"PARAM1"]>1) & 
    (scaled.dat.df[,"PARAM2"]<0 | scaled.dat.df[,"PARAM2"]>1) & 
    (scaled.dat.df[,"PARAM3"]<0 | scaled.dat.df[,"PARAM3"]>1) & 
    (scaled.dat.df[,"PARAM5"]<0 | scaled.dat.df[,"PARAM5"]>1) &
    (scaled.dat.df[,"PARAM6"]<0 | scaled.dat.df[,"PARAM6"]>1) & 
    (scaled.dat.df[,"PARAM7"]<0 | scaled.dat.df[,"PARAM7"]>1) )
noncompliant <- length(z[z == TRUE])

slices <- c(noncompliant, total - noncompliant)
labls <- c("NOT OK","OK")
pct <- round(slices/sum(slices)*100, digits=2)
labls <- paste(labls, pct)
labls <- paste(labls, "%", sep="")

#pie3D(slices,labels=labls,explode=0.05, col=c(rgb(0.75,0,0.5),rgb(0,1,0.75)),main="Defect Analysis due to critical parameters")
pie(slices,labels=labls,main="Defect Analysis due to critical parameters")

#filter non-compliants due to individual params
defect.dat.df <- scaled.dat.df[z,]

#select defect interval range
min <- min(as.numeric(sapply(defect.dat.df[,c("PARAM1","PARAM2","PARAM3","PARAM5","PARAM6","PARAM7")], function(x) min(as.numeric(x)))))
max <- max(as.numeric(sapply(defect.dat.df[,c("PARAM1","PARAM2","PARAM3","PARAM5","PARAM6","PARAM7")], function(x) max(as.numeric(x)))))


#plot histogram for param1 defect
#p1.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM1")], breaks=seq(-0.4,0.2,by=0.2))
p1.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM1")], breaks=seq(min,max,by=0.2))
#h1 <- hist(defect.dat.df[,c("PARAM1")])
#plot(h1, col=rgb(1,0,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h1 <- hist(defect.dat.df[,c("PARAM1")], col=rgb(1,0,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
#h1 <- hist(defect.dat.df[,c("PARAM1")], col=rgb(1,0,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max))
box()

p2.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM2")], breaks=seq(min,max,by=0.2))
#h2 <- hist(defect.dat.df[,c("PARAM2")])
#plot(h2, col=rgb(0,0,1,1/7), xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h2 <- hist(defect.dat.df[,c("PARAM2")], col=rgb(0,0,1,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p2.bin.defect.dat.df$breaks, n=1),tail(p2.bin.defect.dat.df$breaks, n=1)), add=T)
#h2 <- hist(defect.dat.df[,c("PARAM2")], col=rgb(0,0,1,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max), add=T)
box()

p3.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM3")], breaks=seq(min,max,by=0.2))
#h3 <- hist(defect.dat.df[,c("PARAM3")])
#plot(h3, col=rgb(0,1,0,1/7), xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h3 <- hist(defect.dat.df[,c("PARAM3")], col=rgb(0,1,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p3.bin.defect.dat.df$breaks, n=1),tail(p3.bin.defect.dat.df$breaks, n=1)), add=T)
#h3 <- hist(defect.dat.df[,c("PARAM3")], col=rgb(0,1,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max), add=T)
box()

p5.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM5")], breaks=seq(min,max,by=0.2))
#h5 <- hist(defect.dat.df[,c("PARAM5")])
#plot(h5, col=rgb(0.5,0.5,0,1/7), xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h5 <- hist(defect.dat.df[,c("PARAM5")], col=rgb(0.5,0,0.5,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p5.bin.defect.dat.df$breaks, n=1),tail(p5.bin.defect.dat.df$breaks, n=1)), add=T)
#h5 <- hist(defect.dat.df[,c("PARAM5")], col=rgb(0.5,0,0.5,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max), add=T)
box()

p6.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM6")], breaks=seq(min,max,by=0.2))
#h6 <- hist(defect.dat.df[,c("PARAM6")])
#plot(h6, col=rgb(0,0.5,0.5,1/7), xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h6 <- hist(defect.dat.df[,c("PARAM6")], col=rgb(0,0.5,0.5,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p6.bin.defect.dat.df$breaks, n=1),tail(p6.bin.defect.dat.df$breaks, n=1)), add=T)
#h6 <- hist(defect.dat.df[,c("PARAM6")], col=rgb(0,0.5,0.5,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max), add=T)
box()

p7.bin.defect.dat.df <- binning(defect.dat.df[,c("PARAM7")], breaks=seq(min,max,by=0.2))
#h7 <- hist(defect.dat.df[,c("PARAM7")])
#plot(h7, col=rgb(0.5,0,0.5,1/7), xlim=c(head(p1.bin.defect.dat.df$breaks, n=1),tail(p1.bin.defect.dat.df$breaks, n=1)))
h7 <- hist(defect.dat.df[,c("PARAM7")], col=rgb(0.5,0.5,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(head(p7.bin.defect.dat.df$breaks, n=1),tail(p7.bin.defect.dat.df$breaks, n=1)), add=T)
#h7 <- hist(defect.dat.df[,c("PARAM7")], col=rgb(0.5,0.5,0,1/7), xlab="Param Defect Intervals", main="Frequency of Parameter Defects", xlim=c(min,max), add=T)
box()

Answer 1

Really, in order to get the bars the same width you want to use the same breaks across all the groups. I went ahead and created some sample data. Instead of having a bunch of different data.frames, I kept everything in lists.

#sample data
set.seed(15)
observations <- lapply(1:6, function(x) rnorm(100*x))

Here we have 6 items, each with a different number of random draws from a standard normal distribution. Now I'll rind the total range and make 50 breaks

maxrange <- range(sapply(observations, range))
breaks <- seq(maxrange[1], maxrange[2], length.out=50)

Now i'll apply these same breaks to all the sets and i'll calculate the histogram data without actually plotting it

hists <- lapply(bins, hist, breaks=breaks, plot=F)

To help with plotting i need to pre-calculate the the ylim to make sure the first plot is tall enough for all the data for all the other plots. I'll also set up the colors now.

ylim <- range(sapply(hists, function(x) {range(x$density)}))
cols<-list(
    rgb(1,0,0,1/7), rgb(0,0,1,1/7), rgb(0,1,0,1/7),
    rgb(0.5,0,0.5,1/7), rgb(0,0.5,0.5,1/7), rgb(0.5,0.5,0,1/7)
)

Now we plot the first histogram with all the labels and such we want, then we plot all the others on top

plot(hists[[1]], ylim=ylim, col=cols[[1]], freq=F, 
    main="Combined Histogram", xlab="Observation")
invisible(mapply(function(x, c) 
    plot(x, ylim=ylim, col=c, freq=F, add=T), 
hists[-1], cols[-1]))