cubic spline interpolation in R

Question

I am trying an implementation of a cubic spline in R. I have already used the spline, smooth.spline and smooth.Pspline functions that are available in the R libraries but I am not that happy with the results so I want to convince myself about the consistency of the results by a "homemade" spline function. I have already computed the coefficients for the 3rd degree polynomials, but I am not sure how to plot the results..they seem random points. You can find the source code below. Any help would be appreciated.

x = c(35,36,39,42,45,48)
y = c(2.87671519825595, 4.04868309245341,   3.95202175000174,   
  3.87683188946186, 4.07739945984612,   2.16064840967985)


n = length(x)

#determine width of intervals
h=0
for (i in 1:(n-1)){
   h[i] = (x[i+1] - x[i])
}

A = 0
B = 0
C = 0
D = 0
#determine the matrix influence coefficients for the natural spline
for (i in 2:(n-1)){
  j = i-1
  D[j] = 2*(h[i-1] + h[i])
  A[j] = h[i]
  B[j] = h[i-1] 

}

#determine the constant matrix C
for (i in 2:(n-1)){
  j = i-1
  C[j] = 6*((y[i+1] - y[i]) / h[i] - (y[i] - y[i-1]) / h[i-1])
}

#maximum TDMA length
ntdma = n - 2

#tridiagonal matrix algorithm

#upper triangularization
R = 0
for (i in 2:ntdma){
  R = B[i]/D[i-1]
  D[i] = D[i] - R * A[i-1]
  C[i] = C[i] - R * C[i-1] 
}

#set the last C
C[ntdma] = C[ntdma] / D[ntdma]

#back substitute
for (i in (ntdma-1):1){
  C[i] = (C[i] - A[i] * C[i+1]) / D[i]
}

#end of tdma

#switch from C to S
S = 0
for (i in 2:(n-1)){
  j = i - 1
  S[i] = C[j]
}
#end conditions
S[1] <- 0 -> S[n]

#Calculate cubic ai,bi,ci and di from S and h
for (i in 1:(n-1)){
 A[i] = (S[i+ 1] - S[i]) / (6 * h[i])
 B[i] = S[i] / 2
 C[i] = (y[i+1] - y[i]) / h[i] - (2 * h[i] * S[i] + h[i] * S[i + 1]) / 6
 D[i] = y[i]
}


#control points
xx = c(x[2],x[4])
yy = 0
#spline evaluation
for (j in 1:length(xx)){
  for (i in 1:n){
    if (xx[j]<=x[i]){
      break
    }
    yy[i] = A[i]*(xx[j] - x[i])^3 + B[i]*(xx[j] - x[i])^2 + C[i]*(xx[j] - x[i]) + D[i]

 }
points(x,yy ,col="blue")
}

Thank you

Answer 1

Okay here goes...

Your 'control points' here are the points at which you are going to evaluate the cubic spline. So the number of points returned (yy) is the same length as xx. This made me spot something:

for (j in 1:length(xx)){
  for (i in 1:n){
    if (xx[j]<=x[i]){
      break
    }
    yy[i] = A[i]*(xx[j] - x[i])^3 + B[i]*(xx[j] - x[i])^2 + C[i]*(xx[j] - x[i]) + D[i]

 }

This is only computing 'n' values of yy. Hullo, whats wrong here? It should be returning length(xx) values...

Then I think I spotted something else - your 'break' is going to drop out of the for loop. What you really want is to skip that i and go on to the next one until you hit the one relevant to your point:

#spline evaluation
for (j in 1:length(xx)){
  for (i in 1:n){
    if (xx[j]<=x[i]){
      next
     }
    yy[j] = A[i]*(xx[j] - x[i])^3 + B[i]*(xx[j] - x[i])^2 + C[i]*(xx[j] - x[i]) + D[i]

 }
}

This is inefficient because you are computing some yy[j] and dumping them next time round the loop, but no matter, it gets the job done.

Wrap this into a function, so you can play with it easily. My function 'myspline' takes x and y for data to fit, and an xx vector for prediction locations. I can do:

> xx=seq(35,48,len=100)
> yy = myspline(x,y,xx)
> plot(xx,yy,type="l")
> points(x,y)
> 

And I get a nice curve going through the (x,y) points. Except for the first point which it seems to miss and heads off to zero, so I suspect there's still an off-by-one error somewhere. Oh well. 99% done.

Here's the code:

myspline <- function(x,y,xx){

n = length(x)

h=0;yy=0
#determine width of intervals
for (i in 1:(n-1)){
   h[i] = (x[i+1] - x[i])
}

A = 0
B = 0
C = 0
D = 0
#determine the matrix influence coefficients for the natural spline
for (i in 2:(n-1)){
  j = i-1
  D[j] = 2*(h[i-1] + h[i])
  A[j] = h[i]
  B[j] = h[i-1] 

}

#determine the constant matrix C
for (i in 2:(n-1)){
  j = i-1
  C[j] = 6*((y[i+1] - y[i]) / h[i] - (y[i] - y[i-1]) / h[i-1])
}

#maximum TDMA length
ntdma = n - 2

#tridiagonal matrix algorithm

#upper triangularization
R = 0
for (i in 2:ntdma){
  R = B[i]/D[i-1]
  D[i] = D[i] - R * A[i-1]
  C[i] = C[i] - R * C[i-1] 
}

#set the last C
C[ntdma] = C[ntdma] / D[ntdma]

#back substitute
for (i in (ntdma-1):1){
  C[i] = (C[i] - A[i] * C[i+1]) / D[i]
}

#end of tdma

#switch from C to S
S = 0
for (i in 2:(n-1)){
  j = i - 1
  S[i] = C[j]
}
#end conditions
S[1] <- 0 -> S[n]

#Calculate cubic ai,bi,ci and di from S and h
for (i in 1:(n-1)){
 A[i] = (S[i+ 1] - S[i]) / (6 * h[i])
 B[i] = S[i] / 2
 C[i] = (y[i+1] - y[i]) / h[i] - (2 * h[i] * S[i] + h[i] * S[i + 1]) / 6
 D[i] = y[i]
}


#control points
#xx = seq(x[2],x[4],len=100)

#spline evaluation
for (j in 1:length(xx)){
  for (i in 1:n){
    if (xx[j]<=x[i]){
      next
     }
    yy[j] = A[i]*(xx[j] - x[i])^3 + B[i]*(xx[j] - x[i])^2 + C[i]*(xx[j] - x[i]) + D[i]
 }
}
return(yy)
}