在绘制ks软件包的kde.local.test结果时，fhat评估点中的误差单调增加

Question

我很难绘制kde.local.test函数（来自R的ks包）的结果，以二维核密度估计法测试两个样本之间的核密度局部差异。

我准备了两个例子。示例1有效，示例2无效。为什么？

示例1：不起作用

data("iris")  
dat <- iris[order(iris$Sepal.Width, iris$Sepal.Length),]  
x1 <- iris[1:75,c("Sepal.Length", "Sepal.Width")]  
x2 <- iris[76:150,c("Sepal.Length", "Sepal.Width")]  
H1samse <- Hpi(x1, pre = "scale", nstage = 2, pilot = "samse") 
H2samse <- Hpi(x2, pre = "scale", nstage = 2, pilot = "samse")  
loct <- kde.local.test(x1,x2, H1 = H1samse, H2 = H2samse)  
plot(loct)

导致以下错误：

Error in .filled.contour(fhat$eval.points[[1]], fhat$eval.points[[2]],  : 
  invalid contour values: have to increase strictly monotonously

示例2：工作正常

data("crabs")  
x1 <- crabs[crabs$sex=="M", c("FL", "RW")]  
x2 <- crabs[crabs$sex=="F", c("FL", "RW")]  
H1samse <- Hpi(x1, pre = "sphere", nstage = 2, pilot = "samse")  
H2samse <- Hpi(x2, pre = "sphere", nstage = 2, pilot = "samse")  
loct <- kde.local.test(x1,x2, H1 = H1samse, H2 = H2samse)  
plot(loct)  
H1scv<- Hscv(x1)  
H2scv <- Hscv(x2)  
loct <- kde.local.test(x1,x2, H1 = H1scv, H2 = H2scv)  
plot(loct)