dummy-link

ShortStrings

A fast implementation of short strings of fixed size. Great for sorting and group-by operations

Readme

ShortStrings

This is an efficient string format for storing strings using integer types. For example, UInt32 can hold 3 bytes of string with 1 byte to record the size of the string and a UInt128 can hold a 15-byte string with 1 byte to record the size of the string.

Using BitIntegers.jl, integer of larger size than UInt128 can be defined. This package supports string with up to 255 bytes in size.

Quick Start

using ShortStrings

using SortingAlgorithms
using Random: randstring

N = Int(1e6)
svec = [randstring(rand(1:15)) for i=1:N]
# convert to ShortString
ssvec = ShortString15.(svec)

# sort short vectors
@time sort(svec);
@time sort(ssvec, by = x->x.size_content, alg=RadixSort);

# conversion to shorter strings is also possible with
ShortString7(randstring(7))
ShortString3(randstring(3))

# convenience macros are provided for writing actual strings (e.g., for comparison)
s15 = ss15"A short string"  # ShortString15 === ShortString{Int128}
s7 = ss7"shorter"           # ShortString7 === ShortString{Int64}
s3 = ss3"srt"               # ShortString3 === ShortString{Int32}

# The ShortString constructor can automatically select the shortest size that a string will fit in
ShortString("This is a long string")

# The maximum length can also be added:
ShortString("Foo", 15)

# The `ss` macro will also select the shortest size that will fit
s31 = ss"This also is a long string"
0.386383 seconds (126 allocations: 11.450 MiB, 18.62% gc time, 0.59% comp
ilation time)
  0.279547 seconds (742.26 k allocations: 74.320 MiB, 70.85% compilation ti
me)
"This also is a long string"

Benchmarks

using SortingLab, ShortStrings, SortingAlgorithms, BenchmarkTools;
N = Int(1e6);
svec = [randstring(rand(1:15)) for i=1:N];
# convert to ShortString
ssvec = ShortString15.(svec);
basesort = @benchmark sort($svec)
radixsort_timings = @benchmark SortingLab.radixsort($svec)
short_radixsort = @benchmark ShortStrings.fsort($ssvec)
# another way to do sorting
sort(ssvec, by = x->x.size_content, alg=RadixSort)

using RCall
@rput svec;
r_timings = R"""
replicate($(length(short_radixsort.times)), system.time(sort(svec, method="radix"))[3])
""";

using Plots
bar(["Base.sort","SortingLab.radixsort","ShortStrings radix sort", "R radix sort"],
    mean.([basesort.times./1e9, radixsort_timings.times./1e9, short_radixsort.times./1e9, r_timings]),
    title="String sort performance - len: 1m, variable size 15",
    label = "seconds")

using SortingLab, ShortStrings, SortingAlgorithms, BenchmarkTools;
N = Int(1e6);
svec = rand([randstring(rand(1:15)) for i=1:N÷100],N)
# convert to ShortString
ssvec = ShortString15.(svec);
basesort = @benchmark sort($svec) samples = 5 seconds = 120
radixsort_timings = @benchmark SortingLab.radixsort($svec) samples = 5 seconds = 120
short_radixsort = @benchmark ShortStrings.fsort($ssvec) samples = 5 seconds = 120

using RCall

@rput svec;
r_timings = R"""
replicate(max(5, $(length(short_radixsort.times))), system.time(sort(svec, method="radix"))[3])
""";

using Plots
bar(["Base.sort","SortingLab.radixsort","ShortStrings radix sort", "R radix sort"],
    mean.([basesort.times./1e9, radixsort_timings.times./1e9, short_radixsort.times./1e9, r_timings]),
    title="String sort performance - len: $(N÷1_000_000)m, fixed size: 15",
    label = "seconds")

Notes

This is based on the discussion here. If Julia.Base adopts the hybrid representation of strings then it makes this package redundant.

Build Status

contributions welcome

First Commit

01/28/2018

Last Touched

20 days ago

Commits

99 commits

Used By: