Add SmoothedConstantInterpolation

README.md

@@ -48,6 +48,7 @@ corresponding to `(u,t)` pairs.
 
   - `ConstantInterpolation(u,t)` - A piecewise constant interpolation.
 
+  - `SmoothedConstantInterpolation(u,t)` - An integral preserving continuously differentiable approximation of constant interpolation.
   - `LinearInterpolation(u,t)` - A linear interpolation.
   - `QuadraticInterpolation(u,t)` - A quadratic interpolation.
   - `LagrangeInterpolation(u,t,n)` - A Lagrange interpolation of order `n`.

docs/src/index.md

@@ -18,6 +18,7 @@ corresponding to `(u,t)` pairs.
 
   - `ConstantInterpolation(u,t)` - A piecewise constant interpolation.
 
+  - `SmoothedConstantInterpolation(u,t)` - An integral preserving continuously differentiable approximation of constant interpolation.
   - `LinearInterpolation(u,t)` - A linear interpolation.
   - `QuadraticInterpolation(u,t)` - A quadratic interpolation.
   - `LagrangeInterpolation(u,t,n)` - A Lagrange interpolation of order `n`.

docs/src/manual.md

@@ -6,6 +6,7 @@ QuadraticInterpolation
 LagrangeInterpolation
 AkimaInterpolation
 ConstantInterpolation
+SmoothedConstantInterpolation
 QuadraticSpline
 CubicSpline
 BSplineInterpolation

docs/src/methods.md

@@ -77,6 +77,24 @@ A = ConstantInterpolation(u, t, dir = :right)
 plot(A)
 ```
 
+## Smoothed Constant Interpolation
+
+This function is much like the constant interpolation above, but the transition
+between consecutive values is smoothed out so that the function is continuously
+differentiable. The smoothing is done in such a way that the integral of this function
+is never much off from the same integral of constant interpolation without smoothing (because of the symmetry of the smoothing sections).
+The maximum smoothing distance in the `t` direction from the data points can be set
+with the keyword argument `d_max`.
+
+```@example tutorial
+A = ConstantInterpolation(u, t)
+plot(A)
+A = SmoothedConstantInterpolation(u, t; d_max = 10)
+plot!(A)
+```
+
+Note that `u[end]` is ignored.
+
 ## Quadratic Spline
 
 This is the quadratic spline. It is a continuously differentiable interpolation

src/DataInterpolations.jl

@@ -86,9 +86,10 @@ function Base.showerror(io::IO, e::IntegralNotInvertibleError)
 end
 
 export LinearInterpolation, QuadraticInterpolation, LagrangeInterpolation,
-       AkimaInterpolation, ConstantInterpolation, QuadraticSpline, CubicSpline,
-       BSplineInterpolation, BSplineApprox, CubicHermiteSpline, PCHIPInterpolation,
-       QuinticHermiteSpline, LinearInterpolationIntInv, ConstantInterpolationIntInv
+       AkimaInterpolation, ConstantInterpolation, SmoothedConstantInterpolation,
+       QuadraticSpline, CubicSpline, BSplineInterpolation, BSplineApprox,
+       CubicHermiteSpline, PCHIPInterpolation, QuinticHermiteSpline,
+       LinearInterpolationIntInv, ConstantInterpolationIntInv
 
 # added for RegularizationSmooth, JJS 11/27/21
 ### Regularization data smoothing and interpolation

src/derivatives.jl

@@ -19,6 +19,24 @@ function _derivative(A::LinearInterpolation, t::Number, iguess)
     slope
 end
 
+function _derivative(A::SmoothedConstantInterpolation{<:AbstractVector}, t::Number, iguess)
+    idx = get_idx(A, t, iguess)
+    d_lower, d_upper, c_lower, c_upper = get_parameters(A, idx)
+
+    # Fix extrapolation behavior as constant for now
+    if t <= first(A.t) || t >= last(A.t)
+        return zero(c_upper / oneunit(t))
+    end
+
+    if (t - A.t[idx]) < d_lower
+        -2c_lower * ((t - A.t[idx]) / d_lower - 1) / d_lower
+    elseif (A.t[idx + 1] - t) < d_upper
+        2c_upper * (1 - (A.t[idx + 1] - t) / d_upper) / d_upper
+    else
+        zero(c_upper / oneunit(t))
+    end
+end
+
 function _derivative(A::QuadraticInterpolation, t::Number, iguess)
     idx = get_idx(A, t, iguess)
     Δt = t - A.t[idx]

src/integrals.jl

@@ -56,6 +56,38 @@ function _integral(
     end
 end
 
+function _integral(A::SmoothedConstantInterpolation{<:AbstractVector},
+        idx::Number, t1::Number, t2::Number)
+    d_lower, d_upper, c_lower, c_upper = get_parameters(A, idx)
+
+    bound_lower = A.t[idx] + d_lower
+    bound_upper = A.t[idx + 1] - d_upper
+
+    out = A.u[idx] * (t2 - t1)
+
+    # Fix extrapolation behavior as constant for now
+    if t1 <= first(A.t)
+        t1 = first(A.t)
+    elseif t2 >= last(A.t)
+        t2 = last(A.t)
+    end
+
+    if t1 < bound_lower
+        t2_ = min(t2, bound_lower)
+        out -= c_lower * d_lower *
+               (((t2_ - A.t[idx]) / d_lower - 1)^3 - ((t1 - A.t[idx]) / d_lower - 1)^3) / 3
+    end
+
+    if t2 > bound_upper
+        t1_ = max(t1, bound_upper)
+        out += c_upper * d_upper *
+               ((1 - (A.t[idx + 1] - t2) / d_upper)^3 -
+                (1 - (A.t[idx + 1] - t1_) / d_upper)^3) / 3
+    end
+
+    out
+end
+
 function _integral(A::QuadraticInterpolation{<:AbstractVector{<:Number}},
         idx::Number, t1::Number, t2::Number)
     α, β = get_parameters(A, idx)

src/interpolation_caches.jl

@@ -285,6 +285,61 @@ function ConstantInterpolation(
     ConstantInterpolation(u, t, I, dir, extrapolate, cache_parameters, assume_linear_t)
 end
 
+"""
+    SmoothedConstantInterpolation(u, t; d_max = Inf, extrapolate = false,
+        cache_parameters = false, assume_linear_t = 1e-2)
+
+It is a method for interpolating constantly with forward fill, with smoothing around the
+value transitions to make the curve continuously differentiable while the integral never
+drifts far from the integral of constant interpolation. In this interpolation type u[end] is ignored.
+
+## Arguments
+
+  - `u`: data points.
+  - `t`: time points.
+
+## Keyword Arguments
+
+  - `d_max`: the maximum distance in `t` from the data points the smoothing is allowed to reach.
+  - `extrapolate`: boolean value to allow extrapolation. Defaults to `false`.
+  - `cache_parameters`: precompute parameters at initialization for faster interpolation computations. Note: if activated, `u` and `t` should not be modified. Defaults to `false`.
+  - `assume_linear_t`: boolean value to specify a faster index lookup behavior for
+    evenly-distributed abscissae. Alternatively, a numerical threshold may be specified
+    for a test based on the normalized standard deviation of the difference with respect
+    to the straight line (see [`looks_linear`](@ref)). Defaults to 1e-2.
+"""
+struct SmoothedConstantInterpolation{uType, tType, IType, dType, cType, dmaxType, T, N} <:
+       AbstractInterpolation{T, N}
+    u::uType
+    t::tType
+    I::IType
+    p::SmoothedConstantParameterCache{dType, cType}
+    d_max::dmaxType
+    extrapolate::Bool
+    iguesser::Guesser{tType}
+    cache_parameters::Bool
+    linear_lookup::Bool
+    function SmoothedConstantInterpolation(
+            u, t, I, p, d_max, extrapolate, cache_parameters, assume_linear_t)
+        linear_lookup = seems_linear(assume_linear_t, t)
+        N = get_output_dim(u)
+        new{typeof(u), typeof(t), typeof(I), typeof(p.d),
+            typeof(p.c), typeof(d_max), eltype(u), N}(
+            u, t, I, p, d_max, extrapolate, Guesser(t), cache_parameters, linear_lookup)
+    end
+end
+
+function SmoothedConstantInterpolation(u, t; d_max = Inf, extrapolate = false,
+        cache_parameters = false, assume_linear_t = 1e-2)
+    u, t = munge_data(u, t)
+    p = SmoothedConstantParameterCache(u, t, cache_parameters, d_max)
+    A = SmoothedConstantInterpolation(
+        u, t, nothing, p, d_max, extrapolate, cache_parameters, assume_linear_t)
+    I = cumulative_integral(A, cache_parameters)
+    SmoothedConstantInterpolation(
+        u, t, I, p, d_max, extrapolate, cache_parameters, assume_linear_t)
+end
+
 """
     QuadraticSpline(u, t; extrapolate = false, cache_parameters = false)
 

src/interpolation_methods.jl

@@ -116,7 +116,7 @@ function _interpolate(A::AkimaInterpolation{<:AbstractVector}, t::Number, iguess
     @evalpoly wj A.u[idx] A.b[idx] A.c[idx] A.d[idx]
 end
 
-# ConstantInterpolation Interpolation
+# Constant Interpolation
 function _interpolate(A::ConstantInterpolation{<:AbstractVector}, t::Number, iguess)
     if A.dir === :left
         # :left means that value to the left is used for interpolation
@@ -139,6 +139,29 @@ function _interpolate(A::ConstantInterpolation{<:AbstractMatrix}, t::Number, igu
     A.u[:, idx]
 end
 
+# Smoothed constant Interpolation
+function _interpolate(A::SmoothedConstantInterpolation{<:AbstractVector}, t::Number, iguess)
+    idx = get_idx(A, t, iguess)
+    d_lower, d_upper, c_lower, c_upper = get_parameters(A, idx)
+
+    # Fix extrapolation behavior as constant for now
+    if t <= first(A.t)
+        return first(A.u)
+    elseif t >= last(A.t)
+        return A.u[end - 1]
+    end
+
+    out = A.u[idx]
+
+    if (t - A.t[idx]) < d_lower
+        out -= c_lower * ((t - A.t[idx]) / d_lower - 1)^2
+    elseif (A.t[idx + 1] - t) < d_upper
+        out += c_upper * (1 - (A.t[idx + 1] - t) / d_upper)^2
+    end
+
+    out
+end
+
 # QuadraticSpline Interpolation
 function _interpolate(A::QuadraticSpline{<:AbstractVector}, t::Number, iguess)
     idx = get_idx(A, t, iguess)

src/interpolation_utils.jl

@@ -207,6 +207,21 @@ function get_parameters(A::LinearInterpolation, idx)
     end
 end
 
+function get_parameters(A::SmoothedConstantInterpolation, idx)
+    if A.cache_parameters
+        d_lower = A.p.d[idx]
+        d_upper = A.p.d[idx + 1]
+        c_lower = A.p.c[idx]
+        c_upper = A.p.c[idx + 1]
+        d_lower, d_upper, c_lower, c_upper
+    else
+        d_lower, c_lower = smoothed_linear_interpolation_parameters(A.u, A.t, A.d_max, idx)
+        d_upper, c_upper = smoothed_linear_interpolation_parameters(
+            A.u, A.t, A.d_max, idx + 1)
+        d_lower, d_upper, c_lower, c_upper
+    end
+end
+
 function get_parameters(A::QuadraticInterpolation, idx)
     if A.cache_parameters
         A.p.α[idx], A.p.β[idx]

src/parameter_caches.jl

@@ -32,6 +32,31 @@ function linear_interpolation_parameters(u::AbstractArray{T, N}, t, idx) where {
     return slope
 end
 
+struct SmoothedConstantParameterCache{dType, cType}
+    d::dType
+    c::cType
+end
+
+function SmoothedConstantParameterCache(u, t, cache_parameters, d_max)
+    if cache_parameters
+        parameters = smoothed_linear_interpolation_parameters.(
+            Ref(u), Ref(t), d_max, eachindex(t))
+        d, c = collect.(eachrow(stack(collect.(parameters))))
+        SmoothedConstantParameterCache(d, c)
+    else
+        SmoothedConstantParameterCache(eltype(t)[], eltype(u)[])
+    end
+end
+
+function smoothed_linear_interpolation_parameters(u, t, d_max, idx)
+    # TODO: Add support for making periodic extrapolation smooth
+    if isone(idx) || (idx == length(t))
+        zero(one(eltype(t))) / 2, zero(one(eltype(u)) / 2)
+    else
+        min(t[idx] - t[idx - 1], t[idx + 1] - t[idx], 2d_max) / 2, (u[idx] - u[idx - 1]) / 2
+    end
+end
+
 struct QuadraticParameterCache{pType}
     α::pType
     β::pType

test/derivative_tests.jl

@@ -27,7 +27,7 @@ function test_derivatives(method; args = [], kwargs = [], name::String)
         # Interpolation time points
         for _t in t[2:(end - 1)]
             if func isa BSplineInterpolation || func isa BSplineApprox ||
-               func isa CubicHermiteSpline
+               func isa CubicHermiteSpline || func isa SmoothedConstantInterpolation
                 fdiff = forward_fdm(5, 1; geom = true)(func, _t)
                 fdiff2 = forward_fdm(5, 1; geom = true)(t -> derivative(func, t), _t)
             else
@@ -140,6 +140,13 @@ end
     @test all(derivative.(Ref(A), t2 .+ 0.1) .== 0.0)
 end
 
+@testset "SmoothedConstantInterpolation" begin
+    u = [5.5, 2.7, 5.1, 3.0]
+    t = [2.5, 5.6, 6.3, 8.9]
+    test_derivatives(SmoothedConstantInterpolation; args = [u, t],
+        name = "Smoothed constant interpolation")
+end
+
 @testset "Quadratic Spline" begin
     u = [0.0, 1.0, 3.0]
     t = [-1.0, 0.0, 1.0]

test/integral_tests.jl

@@ -65,6 +65,20 @@ end
         name = "Linear Interpolation (Vector) with random points")
 end
 
+@testset "SmoothedConstantInterpolation" begin
+    u = [1.0, 4.0, 9.0, 16.0]
+    t = [1.0, 2.0, 3.0, 4.0]
+    test_integral(
+        SmoothedConstantInterpolation; args = [u, t], name = "Smoothed constant interpolation")
+
+    A_constant = ConstantInterpolation(u, t)
+    I_ref = DataInterpolations.integral(A_constant, first(t), last(t))
+    I_smoothed = [DataInterpolations.integral(
+                      SmoothedConstantInterpolation(u, t; d_max), first(t), last(t))
+                  for d_max in 0.0:0.1:1.0]
+    @test all(I_smoothed .≈ I_ref)
+end
+
 @testset "QuadraticInterpolation" begin
     u = [1.0, 4.0, 9.0, 16.0]
     t = [1.0, 2.0, 3.0, 4.0]

test/interpolation_tests.jl

@@ -501,6 +501,19 @@ end
     @test A(-Inf) == first(u)
 end
 
+@testset "Smoothed constant Interpolation" begin
+    test_interpolation_type(SmoothedConstantInterpolation)
+    u = [0.0, 2.0, 1.0, 3.0]
+    t = [1.2, 2.5, 5.7, 8.7]
+    d_max = 0.5
+    A = SmoothedConstantInterpolation(u, t; d_max)
+    test_cached_index(A)
+
+    @test A(1.9) == u[1]
+    @test A(3.1) == u[2]
+    @test A(2.5) ≈ (u[1] + u[2]) / 2
+end
+
 @testset "QuadraticSpline Interpolation" begin
     test_interpolation_type(QuadraticSpline)
 

test/parameter_tests.jl

@@ -7,6 +7,14 @@ using DataInterpolations
     @test A.p.slope ≈ [4.0, -2.0, 1.0, 0.0]
 end
 
+@testset "Smoothed constant Interpolation" begin
+    u = [1.0, 5.0, 3.0, 4.0, 4.0]
+    t = collect(1:5)
+    A = SmoothedConstantInterpolation(u, t; cache_parameters = true)
+    A.p.d ≈ [0.0, 0.5, 0.5, 0.5, 0.0]
+    A.p.c ≈ [0.0, 2.0, -1.0, 0.5, 0.0]
+end
+
 @testset "Quadratic Interpolation" begin
     u = [1.0, 5.0, 3.0, 4.0, 4.0]
     t = collect(1:5)