This is a Go implementation of the strobemers (minstrobes and randstrobes), with some differences.
The implementation of Randstrobes has a not-bad performance (2-3X slower) compared to regular k-mer,
while it's 10-20X slower than ntHash.
Besides, Randstrobes is only slightly slower than MinStrobes (see benchmark).
The current implementation only computes strobemers of the positive strand, because the strobes are asymmetrical and the location matters.
go get github.com/shenwei356/strobemers
We followed the code style of ntHash.
n := 2
l := 3
w_min := 3
w_max := 5
rs, err := strobemers.NewRandStrobes(seq, n, l, w_min, w_max)
checkError(err)
var hash uint64
var ok bool
var i int // 0-based index
var positions []int // 0-based indexes of all strobes
rs.SetWindowShrink(true)
for {
hash, ok = rs.Next()
if !ok {
break
}
i = rs.Index()
positions = rs.Indexes()
}Here are some differences compared to the original implementation, see discussion: #1, #2.
| item | orginal | this | comment |
|---|---|---|---|
| window range | w_min < w_max |
w_min <= w_max |
allow a fixed position |
| shrinking window | all w_min and w_max |
optional shrinking last w_max |
see figures below |
| number of strobemers | len(seq)-n*l+1 |
len(seq)-n*l+1-(n-1)*l |
window shrinked |
| number of strobemers | len(seq)-n*l+1-(n-1)*(l+w_min-1) |
window not shrinked | |
| choice of min hash | (h(m)+h(mj))%q |
(h(m)+h(mj))&q |
& is faster than % |
| final hash value (n=2) | h(m1)-h(m2) |
h(m1)/2+h(m2)/3 |
keep asymmetry and avoid uint64 overflow |
| final hash value (n=3) | h(m1)-h(m2)+2*h(m3) |
h(m1)/3+h(m2)/4+h(m3)/5 |
~ |
| method | time | relative_time |
|---|---|---|
| ntHashKmers(30) | 8590 | 1 |
| Kmers(30) | 55579 | 6 |
| MinStrobes(2,15,20,30) | 104520 | 12 |
| MinStrobes(3,10,20,30) | 111662 | 13 |
| RandStrobes(2,15,20,30) | 93436 | 11 |
| RandStrobes(3,10,20,30) | 152461 | 18 |
$ go test . -bench=Benchmark* -benchmem \
| grep Bench \
| perl -pe 's/\s\s+/\t/g' \
| csvtk cut -Ht -f 1,3-5 \
| csvtk add-header -t -n test,time,memory,allocs \
| csvtk pretty -t -r
test time memory allocs
------------------------------------- ------------ ---------- -----------
BenchmarkNTHash/1.00_KB-16 8590 ns/op 48 B/op 1 allocs/op
BenchmarkKmers/1.00_KB-16 55579 ns/op 32 B/op 1 allocs/op
BenchmarkMinStrobesOrder2/1.00_KB-16 104520 ns/op 25064 B/op 7 allocs/op
BenchmarkMinStrobesOrder3/1.00_KB-16 111662 ns/op 25064 B/op 7 allocs/op
BenchmarkRandStrobesOrder2/1.00_KB-16 93436 ns/op 8432 B/op 3 allocs/op
BenchmarkRandStrobesOrder3/1.00_KB-16 152461 ns/op 8432 B/op 3 allocs/op