如何生成b样条曲面,比方说:
x=attitude$rating
y=attitude$complaints
z=attitude$privileges
对于样条曲线,将为x和y。 z是一组控制点。
答案 0 :(得分:5)
如果我理解你,你有x,y和z数据,你想在x和y上使用双变量样条插值,使用z作为控制点。您可以使用interp(...)
包中的akima
执行此操作。
library(akima)
spline <- interp(x,y,z,linear=FALSE)
# rotatable 3D plot of points and spline surface
library(rgl)
open3d(scale=c(1/diff(range(x)),1/diff(range(y)),1/diff(range(z))))
with(spline,surface3d(x,y,z,alpha=.2))
points3d(x,y,z)
title3d(xlab="rating",ylab="complaints",zlab="privileges")
axes3d()
图表本身对您的数据集非常不感兴趣,因为x,y和x高度相关。
编辑对OP评论的回应。
如果您需要b样条曲面,请在遗憾的名为mba.surf(...)
的包中试用MBA
。
library(MBA)
spline <- mba.surf(data.frame(x,y,z),100,100)
library(rgl)
open3d(scale=c(1/diff(range(x)),1/diff(range(y)),1/diff(range(z))))
with(spline$xyz,surface3d(x,y,z,alpha=.2))
points3d(x,y,z)
title3d(xlab="rating",ylab="complaints",zlab="privileges")
axes3d()
答案 1 :(得分:5)
require(rms) # Harrell's gift to the R world.
# Better to keep the original names and do so within a dataframe.
att <- attitude[c('rating','complaints','privileges')]
add <- datadist(att) # records ranges and descriptive info on data
options(datadist="add") # need these for the rms functions
# rms-`ols` function (ordinary least squares) is a version of `lm`
mdl <- ols( privileges ~ rcs(rating,4)*rcs(complaints,4) ,data=att)
# Predict is an rms function that works with rms's particular classes
pred <- Predict(mdl, 'rating','complaints')
# bplot calls lattice functions; levelplot by default; this gives a "3d" plot
bplot(pred, yhat~rating+complaints, lfun=wireframe)
它是一个交叉的限制三次样条模型。如果您想要使用您喜欢的样条函数,那么请务必尝试一下。我对rcs
- 函数有好运。
这样可以提供更加开放的网格,计算点数更少:
pred <- Predict(mdl, 'rating','complaints', np=25)
bplot(pred, yhat~rating+complaints, lfun=wireframe)
png()
bplot(pred, yhat~rating+complaints, lfun=wireframe)
dev.off()
您可以使用jhoward所示的rgl方法。 str(pred)的顶部看起来像:
str(pred)
Classes ‘Predict’ and 'data.frame': 625 obs. of 5 variables:
$ rating : num 43 44.6 46.2 47.8 49.4 ...
$ complaints: num 45 45 45 45 45 ...
$ yhat : num 39.9 39.5 39.1 38.7 38.3 ...
$ lower : num 28 28.3 27.3 25 22 ...
$ upper : num 51.7 50.6 50.9 52.4 54.6 ...
snipped
library(rgl)
open3d()
with(pred, surface3d(unique(rating),unique(complaints),yhat,alpha=.2))
with(att, points3d(rating,complaints,privileges, col="red"))
title3d(xlab="rating",ylab="complaints",zlab="privileges")
axes3d()
aspect3d(x=1,z=.05)
一旦你意识到没有关于该模型的不适当推断的极端数据,很好地说明了外推的危险性。 rms-package有一个perimeter
函数,绘图函数有一个perim
参数,周边对象可以传递给它。