sp94 / cubicsplines.jl Goto Github PK

The REQUIRE file could not be found.
cc: @sp94

The REQUIRE file could not be found.
cc: @sp94

Hello,

I'm currently using CubicSplines.jl together with the Unitful.jl package. Since the CubicSpline struct is currently restricted to vectors with real elements, this forces me to convert units to real numbers and back. Could we possibly remove the restriction to real elements (instead restricting it to numbers)?

We should add CI/CD for automatic testing and code coverage

There is the start of a travis.yaml file from #6, but fails on systems without matplotlib installed

• Remove PyPlot as a dependency?
• Migrate to GitHub actions? (Travis has placed limits on free usage)

@StefanMathis thanks very much for your pull request #5. I have a couple questions about some of the documentation

In CubicSplines.jl

The extrapolation behaviour is defined by the coefficients of a 3rd degree
polynomial which can be given as additional keyword arguments extrapl and extrapr
to the constructor. extrapl and/or extrapr should be either "nothing" or a
set of coefficients.

Case 1: Extrapolation is set to "nothing"
In this case, the default quadratic extrapolation according to section 2.3 of [1]
is used to construct the spline and the spline throws an error when evaluated
outside its boundaries.

Case 2: Extrapolation is set to a vector
The container is given as a n-element collection of values:
extrapl = [p1, p2, p3, ..., pn]
which are used to evaluate the following function:
y = p0 + p1*(x-x3) + p2*(x-x3)^2 + p3*(x-x3)^3 + ... + pn*(x-x3)^n
The constant p0 is equal to y3 by definition (continuous spline). x3 and y3
are the first/last datapoint values respectively.

Does the first paragraph contradict the third paragraph? Should we remove "3rd degree" from the first line?

In README.md

The example doesn't run (it doesn't include the xdata and ydata). Do you know what xdata and ydata were used?

Hello Samuel,

I just found that there is no new release of the package yet, which leads to Pkg still downloading the old version w/o extrapolation. Would you mind releasing a new version (v0.1.2)?

I discovered that the interpolated values given from CubicSplines.jl do not match those given by the R function aspline for input values in the interval xs[end-1] to xs[end]

It's not too hard to use a symmetry argument to see that the Julia splined value must be wrong, as set out in the script below. I believe the fix is at line 38 of CubicSplines.jl which should read:
for i in 1:length(ms)
rather than
for i in 1:length(ms)-1

using CubicSplines
using RCall
using Plots

#=CubicSplines2.jl is a copy of CubicSplines.jl with line 38 changed from:
for i in 1:length(ms)-1
to read instead:
for i in 1:length(ms)
=#
include("CubicSplines2.jl")

xs = collect(-2.0:2.0)
ys = mod.(xs, 2)
xin = collect(-2.0:0.5:2.0)

#=By symmetry about zero we should have result(+x) = result(-x)
But that's not true when x is in the interval xs[end-1] to xs[end]

CubicSplines.CubicSpline(xs, ys)[ 1.5] == 0.725
CubicSplines.CubicSpline(xs, ys)[-1.5] == 0.75

As a further check, compare against R's aspline https://www.rdocumentation.org/packages/akima/versions/0.6-2/topics/aspline
=#

j_res = CubicSplines.CubicSpline(xs, ys)[xin]
J_res = CubicSplines2.CubicSpline(xs, ys)[xin]
R_res = rcopy(rcall(:aspline, xs, ys, xin))[:y]

@show xin
@show j_res
@show J_res
@show R_res

@show all(j_res .== R_res)
@show all(J_res .== R_res)

plot(xin,j_res)
plot!(xin,R_res)

running that script:

xin = [-2.0, -1.5, -1.0, -0.5, 0.0, 0.5, 1.0, 1.5, 2.0]
j_res = [0.0, 0.75, 1.0, 0.5, 0.0, 0.5, 1.0, 0.725, 0.0]
J_res = [0.0, 0.75, 1.0, 0.5, 0.0, 0.5, 1.0, 0.75, 0.0]
R_res = [0.0, 0.75, 1.0, 0.5, 0.0, 0.5, 1.0, 0.75, 0.0]
all(j_res .== R_res) = false
all(J_res .== R_res) = true