// Package histogram provides building blocks for fast histograms. package histogram import ( "math" "sort" "sync" "github.com/valyala/fastrand" ) var ( infNeg = math.Inf(-1) infPos = math.Inf(1) nan = math.NaN() ) // Fast is a fast histogram. // // It cannot be used from concurrently running goroutines without // external synchronization. type Fast struct { max float64 min float64 count uint64 a []float64 tmp []float64 rng fastrand.RNG } // NewFast returns new fast histogram. func NewFast() *Fast { f := &Fast{} f.Reset() return f } // Reset resets the histogram. func (f *Fast) Reset() { f.max = infNeg f.min = infPos f.count = 0 f.a = f.a[:0] f.tmp = f.tmp[:0] } // Update updates the f with v. func (f *Fast) Update(v float64) { if v > f.max { f.max = v } if v < f.min { f.min = v } f.count++ if len(f.a) < maxSamples { f.a = append(f.a, v) return } if n := int(f.rng.Uint32n(uint32(f.count))); n < len(f.a) { f.a[n] = v } } const maxSamples = 1000 // Quantile returns the quantile value for the given phi. func (f *Fast) Quantile(phi float64) float64 { f.tmp = append(f.tmp[:0], f.a...) sort.Float64s(f.tmp) return f.quantile(phi) } // Quantiles appends quantile values to dst for the given phis. func (f *Fast) Quantiles(dst, phis []float64) []float64 { f.tmp = append(f.tmp[:0], f.a...) sort.Float64s(f.tmp) for _, phi := range phis { q := f.quantile(phi) dst = append(dst, q) } return dst } func (f *Fast) quantile(phi float64) float64 { if len(f.tmp) == 0 || math.IsNaN(phi) { return nan } if phi <= 0 { return f.min } if phi >= 1 { return f.max } idx := uint(phi*float64(len(f.tmp)-1) + 0.5) if idx >= uint(len(f.tmp)) { idx = uint(len(f.tmp) - 1) } return f.tmp[idx] } // GetFast returns a histogram from a pool. func GetFast() *Fast { v := fastPool.Get() if v == nil { return NewFast() } return v.(*Fast) } // PutFast puts hf to the pool. // // hf cannot be used after this call. func PutFast(f *Fast) { f.Reset() fastPool.Put(f) } var fastPool sync.Pool