Skip to content

Use a faster and safer implementation of alias_sample! #927

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

Merged
merged 3 commits into from
Apr 12, 2024
Merged

Use a faster and safer implementation of alias_sample! #927

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cf44eab
Use a faster implementation for alias_sample!
LilithHafner Apr 8, 2024
b0d1820
add invalid weights tests
LilithHafner Apr 8, 2024
f11375a
deprecate make_alias_table!
LilithHafner Apr 8, 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
2 changes: 2 additions & 0 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ authors = ["JuliaStats"]
version = "0.34.3"

[deps]
AliasTables = "66dad0bd-aa9a-41b7-9441-69ab47430ed8"
DataAPI = "9a962f9c-6df0-11e9-0e5d-c546b8b5ee8a"
DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
Expand All @@ -17,6 +18,7 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
StatsAPI = "82ae8749-77ed-4fe6-ae5f-f523153014b0"

[compat]
AliasTables = "1"
DataAPI = "1"
DataStructures = "0.10, 0.11, 0.12, 0.13, 0.14, 0.17, 0.18"
LinearAlgebra = "<0.0.1, 1"
Expand Down
62 changes: 62 additions & 0 deletions src/deprecates.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,3 +46,65 @@ end
@deprecate stdm(x::AbstractArray{<:Real}, w::AbstractWeights, m::AbstractArray{<:Real}, dim::Int; corrected::Union{Bool, Nothing}=nothing) std(x, w, dim, mean=m, corrected=corrected) false
@deprecate varm(x::AbstractArray{<:Real}, w::AbstractWeights, m::AbstractArray{<:Real}, dim::Int; corrected::Union{Bool, Nothing}=nothing) var(x, w, dim, mean=m, corrected=corrected) false
@deprecate varm!(R::AbstractArray, x::AbstractArray{<:Real}, w::AbstractWeights, m::AbstractArray{<:Real}, dim::Int; corrected::Union{Bool, Nothing}=nothing) var!(R, x, w, dim, mean=m, corrected=corrected) false

### This was never part of the public API
### Deprecated April 2024
function make_alias_table!(w::AbstractVector, wsum,
a::AbstractVector{Float64},
alias::AbstractVector{Int})
Base.depwarn("make_alias_table! is both internal and deprecated, use AliasTables.jl instead", :make_alias_table!)
# Arguments:
#
# w [in]: input weights
# wsum [in]: pre-computed sum(w)
#
# a [out]: acceptance probabilities
# alias [out]: alias table
#
# Note: a and w can be the same array, then that array will be
# overwritten inplace by acceptance probabilities
#
# Returns nothing
#

n = length(w)
length(a) == length(alias) == n ||
throw(DimensionMismatch("Inconsistent array lengths."))

ac = n / wsum
for i = 1:n
@inbounds a[i] = w[i] * ac
end

larges = Vector{Int}(undef, n)
smalls = Vector{Int}(undef, n)
kl = 0 # actual number of larges
ks = 0 # actual number of smalls

for i = 1:n
@inbounds ai = a[i]
if ai > 1.0
larges[kl+=1] = i # push to larges
elseif ai < 1.0
smalls[ks+=1] = i # push to smalls
end
end

while kl > 0 && ks > 0
s = smalls[ks]; ks -= 1 # pop from smalls
l = larges[kl]; kl -= 1 # pop from larges
@inbounds alias[s] = l
@inbounds al = a[l] = (a[l] - 1.0) + a[s]
if al > 1.0
larges[kl+=1] = l # push to larges
else
smalls[ks+=1] = l # push to smalls
end
end

# this loop should be redundant, except for rounding
for i = 1:ks
@inbounds a[smalls[i]] = 1.0
end
nothing
end
82 changes: 9 additions & 73 deletions src/sampling.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
#
###########################################################

using AliasTables
using Random: Sampler

if VERSION < v"1.3.0-DEV.565"
Expand Down Expand Up @@ -635,65 +636,6 @@ end
direct_sample!(a::AbstractArray, wv::AbstractWeights, x::AbstractArray) =
direct_sample!(default_rng(), a, wv, x)

function make_alias_table!(w::AbstractVector, wsum,
a::AbstractVector{Float64},
alias::AbstractVector{Int})
# Arguments:
#
# w [in]: input weights
# wsum [in]: pre-computed sum(w)
#
# a [out]: acceptance probabilities
# alias [out]: alias table
#
# Note: a and w can be the same array, then that array will be
# overwritten inplace by acceptance probabilities
#
# Returns nothing
#

n = length(w)
length(a) == length(alias) == n ||
throw(DimensionMismatch("Inconsistent array lengths."))

ac = n / wsum
for i = 1:n
@inbounds a[i] = w[i] * ac
end

larges = Vector{Int}(undef, n)
smalls = Vector{Int}(undef, n)
kl = 0 # actual number of larges
ks = 0 # actual number of smalls

for i = 1:n
@inbounds ai = a[i]
if ai > 1.0
larges[kl+=1] = i # push to larges
elseif ai < 1.0
smalls[ks+=1] = i # push to smalls
end
end

while kl > 0 && ks > 0
s = smalls[ks]; ks -= 1 # pop from smalls
l = larges[kl]; kl -= 1 # pop from larges
@inbounds alias[s] = l
@inbounds al = a[l] = (a[l] - 1.0) + a[s]
if al > 1.0
larges[kl+=1] = l # push to larges
else
smalls[ks+=1] = l # push to smalls
end
end

# this loop should be redundant, except for rounding
for i = 1:ks
@inbounds a[smalls[i]] = 1.0
end
nothing
end

"""
alias_sample!([rng], a::AbstractArray, wv::AbstractWeights, x::AbstractArray)

Expand All @@ -704,29 +646,23 @@ Build an alias table, and sample therefrom.
Reference: Walker, A. J. "An Efficient Method for Generating Discrete Random Variables
with General Distributions." *ACM Transactions on Mathematical Software* 3 (3): 253, 1977.

Noting `k=length(x)` and `n=length(a)`, this algorithm takes ``O(n \\log n)`` time
for building the alias table, and then ``O(1)`` to draw each sample. It consumes ``2 k`` random numbers.
Noting `k=length(x)` and `n=length(a)`, this algorithm takes ``O(n)`` time
for building the alias table, and then ``O(1)`` to draw each sample. It consumes ``k`` random numbers.
"""
function alias_sample!(rng::AbstractRNG, a::AbstractArray, wv::AbstractWeights, x::AbstractArray)
Base.mightalias(a, x) &&
throw(ArgumentError("output array x must not share memory with input array a"))
Base.mightalias(x, wv) &&
throw(ArgumentError("output array x must not share memory with weights array wv"))
1 == firstindex(a) == firstindex(wv) == firstindex(x) ||
1 == firstindex(a) == firstindex(wv) ||
throw(ArgumentError("non 1-based arrays are not supported"))
n = length(a)
length(wv) == n || throw(DimensionMismatch("Inconsistent lengths."))
length(wv) == length(a) || throw(DimensionMismatch("Inconsistent lengths."))

# create alias table
ap = Vector{Float64}(undef, n)
alias = Vector{Int}(undef, n)
make_alias_table!(wv, sum(wv), ap, alias)
at = AliasTable(wv)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I guess there is no reason to keep make_alias_table! (defined above) around.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

According to the public API of StatsBase, no, there's no reason. However there are some usages downstream of that internal function. Most notably, Distributions.jl, which would break if we remove make_alias_table! before we merge JuliaStats/Distributions.jl#1848. Once Distributions.jl no longer depends on this, I think it's worth removing.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems Distributions might be the only more widely used package that relies on these internals? I was aware of potential problems with using an internal function when writing ConsistencyResampling, and would be fine with breaking it (and probably switching to AliasTables) 😛 But given the impact on Distributions, maybe it would be safer to deprecate make_alias_table! and remove it in a future breaking release.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

IMO it would even be acceptable to deprecate it and remove it in a future non-breaking release once Distributions has a few releases out that don't use it.


# sampling
s = Sampler(rng, 1:n)
for i = 1:length(x)
j = rand(rng, s)
x[i] = rand(rng) < ap[j] ? a[j] : a[alias[j]]
for i in eachindex(x)
j = rand(rng, at)
x[i] = a[j]
end
return x
end
Expand Down
3 changes: 3 additions & 0 deletions test/wsampling.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,9 @@ for wv in (
check_wsample_wrep(a, (4, 7), wv, 5.0e-3; ordered=false)
end

@test_throws ArgumentError alias_sample!(rand(10), weights(fill(0, 10)), rand(10))
@test_throws ArgumentError alias_sample!(rand(100), weights(randn(100)), rand(10))

for rev in (true, false), T in (Int, Int16, Float64, Float16, BigInt, ComplexF64, Rational{Int})
r = rev ? reverse(4:7) : (4:7)
r = T===Int ? r : T.(r)
Expand Down