Merge pull request #7 from xiaodaigh/missing

support sorting of missing

Author: xiaodaigh

NEWS.md

@@ -1,3 +1,7 @@
+## Version 0.2.2
+* Supports `sort` `Union{T, Missing}`
+* Supports `fsortperm` of `Union{T, Missing}`
+
 ## Version 0.2.1
 
 * fsortperm now sorts many more types including Floats and Int

Project.toml

@@ -1,18 +1,18 @@
 name = "SortingLab"
 uuid = "562c1548-17b8-5b69-83cf-d8aebec229f5"
 authors = ["Dai ZJ <zhuojia.dai@gmail.com>"]
-version = "0.2.1"
+version = "0.2.2"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
 SortingAlgorithms = "a2af1166-a08f-5f64-846c-94a0d3cef48c"
 StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 
 [compat]
-julia = "1"
 CategoricalArrays = "0.7"
 SortingAlgorithms = "0.3"
 StatsBase = "0.32"
+julia = "1"
 
 [extras]
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"

README.md

@@ -78,8 +78,6 @@ Test Passed
 
 
 
-
-
 ## Benchmark
 ![Base.sort vs SortingLab.radixsort](benchmarks/sort_vs_radixsort.png)
 

src/SortingLab.jl

@@ -9,7 +9,6 @@ module SortingLab
 # import InternedStrings
 import StatsBase: countmap
 import CategoricalArrays: CategoricalArray
-import SortingAlgorithms: uint_mapping
 import Base.Threads: @threads
 
 export fsortperm, radixsort, radixsort!, fsort, fsort!
@@ -24,6 +23,7 @@ include("fsortperm_string.jl")
 include("fsortperm_Integer.jl")
 include("fsort_CategoricalArrays.jl")
 include("fsort.jl")
+include("fsort-missing.jl")
 # include("radixsort-StrFs.jl")
 # include("../benchmarks/is_parallel_hist_faster_YES.jl")
 end # module

src/fsort-missing.jl

@@ -0,0 +1,7 @@
+fsort(x::AbstractVector{Union{T, Missing}}; rev = false) where T = begin
+    nmissing = mapreduce(ismissing, +, x)
+    cx = similar(x)
+    cx[1:length(x) - nmissing] .= fsort(collect(skipmissing(x)); rev = rev)
+    cx[length(x)+1:end] .= missing
+    cx
+end

src/fsortperm-missing.jl

@@ -0,0 +1,6 @@
+fsortperm(x::AbstractVector{Union{Missing, T}}) where T =  begin
+    n_missing = mapreduce(ismissing, +, x)
+    res = Vector{Int}(undef, length(x))
+    res[1:length(x)-n_missing] .= fsortperm(collect(skipmissing(x)))
+    
+end

src/sorttwo!.jl

@@ -6,7 +6,7 @@ import StatsBase: BaseRadixSortSafeTypes
 Sort both the `vs` and reorder `index` at the same. This allows for faster sortperm
 for radix sort.
 """
-function sorttwo!(vs::Vector{T}, index, lo::Int = 1, hi::Int=length(vs), RADIX_SIZE = 16, RADIX_MASK = 0xffff) where T <:Union{BaseRadixSortSafeTypes}
+function sorttwo!(vs::Vector{T}, index, lo::Int = 1, hi::Int=length(vs), RADIX_SIZE = 16, RADIX_MASK = 0xffff) where T # <:Union{BaseRadixSortSafeTypes}
     # Input checking
     if lo >= hi;  return (vs, index);  end
     #println(vs)

src/uint_hist.jl

@@ -1,6 +1,7 @@
 # winner from benchmarks/which_is_the_fastest_UInt_histogram.jl
 """
     uint_hist(bits, [RADIX_SIZE = 16, RADIX_MASK = 0xffff])
+    
 Computes a histogram (counts) for the vector RADIX_SIZE bits at a time. E.g. if eltype(bits) is UInt64 and RADIX_SIZE is 16
 then 4 histograms are created for each of the 16 bit chunks.
 """

src/uint_mapping.jl

@@ -1,3 +1,9 @@
+export uint_mapping
+
 import SortingAlgorithms: uint_mapping
 
 uint_mapping(x) = uint_mapping(Base.Forward, x)
+
+uint_mapping(::Missing) = UInt(2)^8sizeof(UInt)-1
+
+uint_mapping(_, ::Missing) = UInt(2)^8sizeof(UInt)-1

test/fsort-missing.jl

@@ -0,0 +1,9 @@
+using Test
+using SortingLab
+
+x = [1,2,missing, 3]
+
+SortingLab.sorttwo!(x, collect(1:4))
+fsortperm(x)
+
+@test isequal(sort(x), fsort(x))