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

Remove LoopVectorization dependency, in light of deprecation #48

Merged
merged 8 commits into from
Jun 13, 2024
Merged

Remove LoopVectorization dependency, in light of deprecation #48

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
d0c962c
De-lv ArrayStats.jl
brenhinkeller Jun 12, 2024
ea59d26
Improve dimensionality-checking in ArrayStats functions
brenhinkeller Jun 12, 2024
8ac5d8d
De-lv other summary statistics
brenhinkeller Jun 13, 2024
359b367
De-lv `quicksort`
brenhinkeller Jun 13, 2024
a815f87
Relax type constraints in `nanvar`
brenhinkeller Jun 13, 2024
29ed673
Remove LoopVectorization dependency
brenhinkeller Jun 13, 2024
b1c2ca7
Don't assume that `partialsort!` partitions around the selected index
brenhinkeller Jun 13, 2024
0af934c
Bump version to 0.6.36
brenhinkeller Jun 13, 2024
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
12 changes: 4 additions & 8 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,19 +1,15 @@
name = "NaNStatistics"
uuid = "b946abbf-3ea7-4610-9019-9858bfdeaf2d"
authors = ["C. Brenhin Keller"]
version = "0.6.35"
version = "0.6.36"

[deps]
IfElse = "615f187c-cbe4-4ef1-ba3b-2fcf58d6d173"
LoopVectorization = "bdcacae8-1622-11e9-2a5c-532679323890"
PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
StaticArrayInterface = "0d7ed370-da01-4f52-bd93-41d350b8b718"

[compat]
IfElse = "0.1"
LoopVectorization = "0.12.113"
VectorizationBase = "0.21.0 - 0.21.64"
PrecompileTools = "1"
Static = "0.8"
julia = "1.8"
StaticArrayInterface = "1"
julia = "1.10"
114 changes: 24 additions & 90 deletions src/ArrayStats/ArrayStats.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,16 +11,9 @@
```
Return the number of elements of `A` that are `NaN`s.
"""
function countnans(A::StridedArray{T}) where T<:PrimitiveNumber
n = 0
@turbo check_empty=true for i ∈ eachindex(A)
n += A[i]!=A[i]
end
return n
end
function countnans(A)
n = 0
@inbounds for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A)
n += A[i]!=A[i]
end
return n
Expand All @@ -35,7 +28,7 @@
"""
function countnotnans(A)
n = 0
@turbo check_empty=true for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A)
n += A[i]==A[i]
end
return n
Expand All @@ -57,14 +50,8 @@
```
Fill a Boolean mask of dimensions `size(A)` that is false wherever `A` is `NaN`
"""
function nanmask!(mask::StridedArray, A::StridedArray{T}) where T<:PrimitiveNumber
@turbo for i ∈ eachindex(A)
mask[i] = A[i]==A[i]
end
return mask
end
function nanmask!(mask, A)
@inbounds for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A)
mask[i] = A[i]==A[i]
end
return mask
Expand All @@ -81,21 +68,11 @@
```
Replace all `NaN`s in A with zeros of the same type
"""
function zeronan!(A::StridedArray{T}) where T<:PrimitiveNumber
∅ = zero(T)
@turbo for i ∈ eachindex(A)
Aᵢ = A[i]
A[i] = ifelse(Aᵢ==Aᵢ, Aᵢ, ∅)
end
return A
end
function zeronan!(A::AbstractArray{T}) where T
∅ = zero(T)
@inbounds for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A)
Aᵢ = A[i]
if isnan(Aᵢ)
A[i] = ∅
end
A[i] = ifelse(Aᵢ==Aᵢ, Aᵢ, ∅)
end
return A
end
Expand Down Expand Up @@ -153,9 +130,8 @@
function nanadd(A::AbstractArray, B::AbstractArray)
result_type = promote_type(eltype(A), eltype(B))
result = similar(A, result_type)
@inbounds @simd for i ∈ eachindex(A)
Aᵢ = A[i]
Bᵢ = B[i]
@inbounds @simd ivdep for i ∈ eachindex(A,B)
Aᵢ, Bᵢ = A[i], B[i]
result[i] = (Aᵢ * (Aᵢ==Aᵢ)) + (Bᵢ * (Bᵢ==Bᵢ))
end
return result
Expand All @@ -169,9 +145,8 @@
Add the non-NaN elements of `B` to `A`, treating `NaN`s as zeros
"""
function nanadd!(A::Array, B::AbstractArray)
@inbounds @simd for i ∈ eachindex(A)
Aᵢ = A[i]
Bᵢ = B[i]
@inbounds @simd for i ∈ eachindex(A,B)
Aᵢ, Bᵢ = A[i], B[i]
A[i] = (Aᵢ * (Aᵢ==Aᵢ)) + (Bᵢ * (Bᵢ==Bᵢ))
end
return A
Expand All @@ -198,7 +173,6 @@
__nanminimum(A, ::Colon, region) = reducedims(_nanminimum(A, region), region)
_nanminimum(A, region) = reduce(nanmin, A, dims=region, init=float(eltype(A))(NaN))
_nanminimum(A, ::Colon) = reduce(nanmin, A, init=float(eltype(A))(NaN))
_nanminimum(A::Array{<:Number}, ::Colon) = vreduce(nanmin, A)
export nanminimum


Expand All @@ -219,7 +193,6 @@
__nanmaximum(A, ::Colon, region) = reducedims(_nanmaximum(A, region), region)
_nanmaximum(A, region) = reduce(nanmax, A, dims=region, init=float(eltype(A))(NaN))
_nanmaximum(A, ::Colon) = reduce(nanmax, A, init=float(eltype(A))(NaN))
_nanmaximum(A::Array{<:Number}, ::Colon) = vreduce(nanmax, A)
export nanmaximum

"""
Expand Down Expand Up @@ -312,41 +285,25 @@
mask = nanmask(A)
return sum(A.*W.*mask, dims=region) ./ sum(W.*mask, dims=region)
end
# Fallback method for non-StridedArrays
function _nanmean(A, W, ::Colon)
n = zero(eltype(W))
m = zero(promote_type(eltype(W), eltype(A)))
@inbounds @simd for i ∈ eachindex(A)
Aᵢ = A[i]
Wᵢ = W[i]
t = Aᵢ == Aᵢ
n += Wᵢ * t
m += Wᵢ * Aᵢ * t
end
return m / n
end
# Can't have NaNs if array is all Integers
function _nanmean(A::StridedArray{<:Integer}, W, ::Colon)
function _nanmean(A::AbstractArray{<:Integer}, W, ::Colon)
n = zero(eltype(W))
m = zero(promote_type(eltype(W), eltype(A)))
@turbo for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A,W)
Wᵢ = W[i]
n += Wᵢ
m += Wᵢ * A[i]
end
return m / n
end
function _nanmean(A::StridedArray, W, ::Colon)
T1 = eltype(W)
T2 = promote_type(eltype(W), eltype(A))
n = zero(T1)
m = zero(T2)
@turbo for i ∈ eachindex(A)
Aᵢ = A[i]
Wᵢ = W[i]
function _nanmean(A, W, ::Colon)
n = ∅ₙ = zero(eltype(W))
m = ∅ₘ = zero(promote_type(eltype(W), eltype(A)))
@inbounds @simd ivdep for i ∈ eachindex(A,W)
Aᵢ, Wᵢ = A[i], W[i]
t = Aᵢ==Aᵢ
n += ifelse(t, Wᵢ, zero(T1))
m += ifelse(t, Wᵢ * Aᵢ, zero(T2))
n += ifelse(t, Wᵢ, ∅ₙ)
m += ifelse(t, Wᵢ * Aᵢ, ∅ₘ)
end
return m / n
end
Expand Down Expand Up @@ -374,60 +331,37 @@
w = sum(W.*mask, dims=region)
s = sum(A.*W.*mask, dims=region) ./ w
d = A .- s # Subtract mean, using broadcasting
@turbo for i ∈ eachindex(d)
@inbounds @simd ivdep for i ∈ eachindex(d, W)
dᵢ = d[i]
d[i] = (dᵢ * dᵢ * W[i]) * mask[i]
end
s .= sum(d, dims=region)
@turbo for i ∈ eachindex(s)
@inbounds @simd ivdep for i ∈ eachindex(s,n,w)
s[i] = sqrt((s[i] * n[i]) / (w[i] * (n[i] - 1)))
end
return s
end
function _nanstd(A, W, ::Colon)
@assert eachindex(A) == eachindex(W)
n = 0
w = zero(eltype(W))
m = zero(promote_type(eltype(W), eltype(A)))
@inbounds @simd for i ∈ eachindex(A)
Aᵢ = A[i]
Wᵢ = W[i]
@inbounds @simd ivdep for i ∈ eachindex(A,W)
Aᵢ, Wᵢ = A[i], W[i]
t = Aᵢ == Aᵢ
n += t
w += Wᵢ * t
m += Wᵢ * Aᵢ * t
end
mu = m / w
s = zero(typeof(mu))
@inbounds @simd for i ∈ eachindex(A)
@inbounds @simd ivdep for i ∈ eachindex(A,W)
Aᵢ = A[i]
d = Aᵢ - mu
s += (d * d * W[i]) * (Aᵢ == Aᵢ) # Zero if Aᵢ is NaN
end
return sqrt(s / w * n / (n-1))
end
function _nanstd(A::StridedArray{Ta}, W::StridedArray{Tw}, ::Colon) where {Ta<:PrimitiveNumber, Tw<:PrimitiveNumber}
n = 0
Tm = promote_type(Tw, Ta)
w = zero(Tw)
m = zero(Tm)
@turbo for i ∈ eachindex(A)
Aᵢ = A[i]
Wᵢ = W[i]
t = Aᵢ==Aᵢ
n += t
w += ifelse(t, Wᵢ, zero(Tw))
m += ifelse(t, Wᵢ * Aᵢ, zero(Tm))
end
mu = m / w
Tmu = typeof(mu)
s = zero(Tmu)
@turbo for i ∈ eachindex(A)
Aᵢ = A[i]
d = Aᵢ - mu
s += ifelse(Aᵢ==Aᵢ, d * d * W[i], zero(Tmu))
end
return sqrt(s / w * n / (n-1))
end
export nanstd


Expand Down
50 changes: 7 additions & 43 deletions src/ArrayStats/nancov.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@ function _nancov(x::AbstractVector, y::AbstractVector, corrected::Bool, μᵪ::N
# Calculate covariance
σᵪᵧ = ∅ = zero(promote_type(typeof(μᵪ), typeof(μᵧ), Int))
n = 0
@turbo check_empty=true for i ∈ eachindex(x,y)
@inbounds @simd ivdep for i ∈ eachindex(x,y)
δᵪ = x[i] - μᵪ
δᵧ = y[i] - μᵧ
δ² = δᵪ * δᵧ
Expand All @@ -19,7 +19,7 @@ function _nancov(x::AbstractVector, y::AbstractVector, corrected::Bool)
n = 0
Σᵪ = ∅ᵪ = zero(eltype(x))
Σᵧ = ∅ᵧ = zero(eltype(y))
@turbo check_empty=true for i ∈ eachindex(x,y)
@inbounds @simd ivdep for i ∈ eachindex(x,y)
xᵢ, yᵢ = x[i], y[i]
notnan = (xᵢ==xᵢ) & (yᵢ==yᵢ)
n += notnan
Expand Down Expand Up @@ -103,49 +103,13 @@ function nancor(x::AbstractVector, y::AbstractVector; corrected::Bool=true)
@assert eachindex(x) == eachindex(y)
return _nancor(x, y, corrected)
end

# Pair-wise nan-covariance
function _nancor(x::StridedVector{T}, y::StridedVector{T}, corrected::Bool) where T<:PrimitiveNumber
function _nancor(x::AbstractVector{Tx}, y::AbstractVector{Ty}, corrected::Bool) where {Tx, Ty}
# Parwise nan-means
n = 0
Σᵪ = ∅ᵪ = zero(T)
Σᵧ = ∅ᵧ = zero(T)
@turbo check_empty=true for i ∈ eachindex(x,y)
xᵢ, yᵢ = x[i], y[i]
notnan = (xᵢ==xᵢ) & (yᵢ==yᵢ)
n += notnan
Σᵪ += ifelse(notnan, xᵢ, ∅ᵪ)
Σᵧ += ifelse(notnan, yᵢ, ∅ᵧ)
end
μᵪ = Σᵪ/n
μᵧ = Σᵧ/n
n == 0 && return (∅ᵪ+∅ᵧ)/0 # Return appropriate NaN if no data

# Pairwise nan-variances
σ²ᵪ = ∅ᵪ = zero(typeof(μᵪ))
σ²ᵧ = ∅ᵧ = zero(typeof(μᵧ))
@turbo check_empty=true for i ∈ eachindex(x,y)
δᵪ = x[i] - μᵪ
δᵧ = y[i] - μᵧ
notnan = (δᵪ==δᵪ) & (δᵧ==δᵧ)
σ²ᵪ += ifelse(notnan, δᵪ * δᵪ, ∅ᵪ)
σ²ᵧ += ifelse(notnan, δᵧ * δᵧ, ∅ᵧ)
end
σᵪ = sqrt(σ²ᵪ / max(n-corrected, 0))
σᵧ = sqrt(σ²ᵧ / max(n-corrected, 0))

# Covariance and correlation
σᵪᵧ = _nancov(x, y, corrected, μᵪ, μᵧ)
ρᵪᵧ = σᵪᵧ / (σᵪ * σᵧ)

return ρᵪᵧ
end
function _nancor(x::AbstractVector, y::AbstractVector, corrected::Bool)
# Parwise nan-means
n = 0
Σᵪ = ∅ᵪ = zero(eltype(x))
Σᵧ = ∅ᵧ = zero(eltype(y))
@inbounds for i ∈ eachindex(x,y)
Σᵪ = ∅ᵪ = zero(Tx)
Σᵧ = ∅ᵧ = zero(Ty)
@inbounds @simd ivdep for i ∈ eachindex(x,y)
xᵢ, yᵢ = x[i], y[i]
notnan = (xᵢ==xᵢ) & (yᵢ==yᵢ)
n += notnan
Expand All @@ -159,7 +123,7 @@ function _nancor(x::AbstractVector, y::AbstractVector, corrected::Bool)
# Pairwise nan-variances
σ²ᵪ = ∅ᵪ = zero(typeof(μᵪ))
σ²ᵧ = ∅ᵧ = zero(typeof(μᵧ))
@inbounds for i ∈ eachindex(x,y)
@inbounds @simd ivdep for i ∈ eachindex(x,y)
δᵪ = x[i] - μᵪ
δᵧ = y[i] - μᵧ
notnan = (δᵪ==δᵪ) & (δᵧ==δᵧ)
Expand Down
Loading
Loading