Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

add optional scaling to ShiftedExp #96

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

add optional scaling to ShiftedExp #96

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 29 additions & 19 deletions src/scalar.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,32 +54,33 @@ inverse(::Identity, x::Real) = x
"""
$(TYPEDEF)

Shifted exponential. When `D::Bool == true`, maps to `(shift, ∞)` using `x ↦
shift + eˣ`, otherwise to `(-∞, shift)` using `x ↦ shift - eˣ`.
Shifted exponential. When `D::Bool == true`, maps to `(shift, scale, ∞)` using `x ↦
shift + exp(x/scale)`, otherwise to `(-∞, shift)` using `x ↦ shift - exp(x/scale)`.
"""
struct ShiftedExp{D, T <: Real} <: ScalarTransform
struct ShiftedExp{D, T <: Real, S <: Real} <: ScalarTransform
shift::T
function ShiftedExp{D,T}(shift::T) where {D, T <: Real}
scale::S
function ShiftedExp{D,T,S}(shift::T, scale::S) where {D, T <: Real, S <: Real}
@argcheck D isa Bool
new(shift)
new(shift, scale)
end
end

ShiftedExp(D::Bool, shift::T) where {T <: Real} = ShiftedExp{D,T}(shift)
ShiftedExp(D::Bool, shift::T) where {T <: Real} = ShiftedExp{D,T,T}(shift, one(T))
ShiftedExp(D::Bool, shift::T, scale::S) where {T <: Real, S <: Real} = ShiftedExp{D,T,S}(shift, scale)

transform(t::ShiftedExp{D}, x::Real) where D =
D ? t.shift + exp(x) : t.shift - exp(x)
D ? t.shift + exp(x/t.scale) : t.shift - exp(x/t.scale)

transform_and_logjac(t::ShiftedExp, x::Real) = transform(t, x), x
transform_and_logjac(t::ShiftedExp, x::Real) = transform(t, x), x/t.scale - log(t.scale)

function inverse(t::ShiftedExp{D}, x::Real) where D
@unpack shift = t
@unpack shift, scale = t
if D
@argcheck x > shift DomainError
log(x - shift)
scale*log(x - shift)
else
@argcheck x < shift DomainError
log(shift - x)
scale*log(shift - x)
end
end

Expand Down Expand Up @@ -155,23 +156,32 @@ Return a transformation that transforms a single real number to the given (open)
interval.

`left < right` is required, but may be `-∞` or `∞`, respectively, in which case
the appropriate transformation is selected. See [`∞`](@ref).
the appropriate transformation is selected. See [`∞`](@ref). If `left` or `right`
are infinite, optionally the scale of the variable can be provied, e.g.:
```
as(Real, left, ∞; scale=10)
````

Some common transformations are predefined as constants, see [`asℝ`](@ref),
[`asℝ₋`](@ref), [`asℝ₊`](@ref), [`as𝕀`](@ref).

!!! note
The finite arguments are promoted to a common type and affect promotion. Eg
`transform(as(0, ∞), 0f0) isa Float32`, but `transform(as(0.0, ∞), 0f0) isa Float64`.
The finite arguments are promoted to a common type and affect promotion. E.g.
`transform(as(0, ∞; scale=10f0), 0f0) isa Float32`, but
`transform(as(0.0, ∞), 0f0) isa Float64`.
"""
as(::Type{Real}, left, right) =
throw(ArgumentError("($(left), $(right)) must be an interval"))

as(::Type{Real}, ::Infinity{false}, ::Infinity{true}) = Identity()

as(::Type{Real}, left::Real, ::Infinity{true}) = ShiftedExp(true, left)
function as(::Type{Real}, left::T, ::Infinity{true}; scale=1) where T <: Real
ShiftedExp(true, left, scale)
end

as(::Type{Real}, ::Infinity{false}, right::Real) = ShiftedExp(false, right)
function as(::Type{Real}, ::Infinity{false}, right::T; scale=1) where T <: Real
ShiftedExp(false, right, scale)
end

function as(::Type{Real}, left::Real, right::Real)
@argcheck left < right "the interval ($(left), $(right)) is empty"
Expand Down Expand Up @@ -220,9 +230,9 @@ Base.show(io::IO, t::ShiftedExp) =
elseif t === asℝ₋
print(io, "asℝ₋")
elseif t isa ShiftedExp{true}
print(io, "as(Real, ", t.shift, ", ∞)")
print(io, "as(Real, ", t.shift, ", ∞; scale = ", t.scale, ")")
else
print(io, "as(Real, -∞, ", t.shift, ")")
print(io, "as(Real, -∞, ", t.shift, "; scale = ", t.scale, ")")
end

Base.show(io::IO, t::ScaledShiftedLogistic) =
Expand Down
27 changes: 21 additions & 6 deletions test/runtests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,15 @@ end
a = randn() * 100
test_transformation(as(Real, -∞, a), y -> y < a)
test_transformation(as(Real, a, ∞), y -> y > a)

a = randn() * 1000
test_transformation(as(Real, a, ∞, scale = 10.0), y -> y > a)
test_transformation(as(Real, -∞, a, scale = 10.0), y -> y < a)

a = randn() * 10
test_transformation(as(Real, a, ∞, scale = 0.1), y -> y > a)
test_transformation(as(Real, -∞, a, scale = 0.1), y -> y < a)

b = a + 0.5 + rand(Float64) + exp(randn() * 10)
test_transformation(as(Real, a, b), y -> a < y < b)
end
Expand All @@ -61,9 +70,13 @@ end

t = as(Real, 1.0, ∞)
@test_throws DomainError inverse(t, 0.5)
t = as(Real, 1.0, ∞; scale = 0.1)
@test_throws DomainError inverse(t, 0.5)

t = as(Real, -∞, 10.0)
@test_throws DomainError inverse(t, 11.0)
t = as(Real, -∞, 10.0; scale=0.1)
@test_throws DomainError inverse(t, 11.0)

t = as(Real, 1.0, 10.0)
@test_throws DomainError inverse(t, 0.5)
Expand All @@ -85,6 +98,8 @@ end
@test transform(asℝ₊, a) isa Float32
@test transform(asℝ₋, a) isa Float32
@test transform(as𝕀, a) isa Float32
@test transform(as(Real, 0, ∞), a) isa Float32
@test transform(as(Real, -∞, 0), a) isa Float32
end

####
Expand Down Expand Up @@ -529,24 +544,24 @@ end
@test string(asℝ₋) == "asℝ₋"
@test string(as𝕀) == "as𝕀"
@test string(as(Real, 0.0, 2.0)) == "as(Real, 0.0, 2.0)"
@test string(as(Real, 1.0, ∞)) == "as(Real, 1.0, ∞)"
@test string(as(Real, -∞, 1.0)) == "as(Real, -∞, 1.0)"
@test string(as(Real, 1.0, ∞)) == "as(Real, 1.0, ∞; scale = 1)"
@test string(as(Real, -∞, 1.0)) == "as(Real, -∞, 1.0; scale = 1)"
end

@testset "sum dimensions allocations" begin
shifted = TransformVariables.ShiftedExp{true,Float64}(0.0)
shifted = TransformVariables.ShiftedExp{true,Float64,Float64}(0.0, 0.0)
tr = (a = shifted, b = TransformVariables.Identity(), c = shifted, d = shifted, e = shifted, f = shifted)
@test iszero(@allocated TransformVariables._sum_dimensions(tr))
end

if VERSION >= v"1.7"
@testset "inverse_eltype allocations" begin
trf = as((x0 = TransformVariables.ShiftedExp{true, Float32}(0f0), x1 = TransformVariables.Identity(), x2 = UnitSimplex(7), x3 = TransformVariables.CorrCholeskyFactor(5), x4 = as(Real, -∞, 1), x5 = as(Array, 10, 2), x6 = as(Array, as𝕀, 10), x7 = as((a = asℝ₊, b = as𝕀)), x8 = TransformVariables.UnitVector(10), x9 = TransformVariables.ShiftedExp{true, Float32}(0f0), x10 = TransformVariables.ShiftedExp{true, Float32}(0f0), x11 = TransformVariables.ShiftedExp{true, Float32}(0f0), x12 = TransformVariables.ShiftedExp{true, Float32}(0f0), x13 = TransformVariables.Identity(), x14 = TransformVariables.ShiftedExp{true, Float32}(0f0), x15 = TransformVariables.ShiftedExp{true, Float32}(0f0), x16 = TransformVariables.ShiftedExp{true, Float32}(0f0), x17 = TransformVariables.ShiftedExp{true, Float64}(0.0)));
trf = as((x0 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x1 = TransformVariables.Identity(), x2 = UnitSimplex(7), x3 = TransformVariables.CorrCholeskyFactor(5), x4 = as(Real, -∞, 1), x5 = as(Array, 10, 2), x6 = as(Array, as𝕀, 10), x7 = as((a = asℝ₊, b = as𝕀)), x8 = TransformVariables.UnitVector(10), x9 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x10 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x11 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x12 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x13 = TransformVariables.Identity(), x14 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x15 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x16 = TransformVariables.ShiftedExp{true, Float32, Float32}(0f0, 0f0), x17 = TransformVariables.ShiftedExp{true, Float64, Float64}(0.0, 0.0)));

vx = randn(@inferred(TransformVariables.dimension(trf)));
x = TransformVariables.transform(trf, vx);
@test @inferred(TransformVariables.inverse_eltype(trf, x)) === Float64

end
end

Expand Down