cerc-io/lotus
16
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 27 Wiki Activity

Change alg for computing variance

Browse Source 
Signed-off-by: Jakub Sztandera <kubuxu@protocol.ai>
This commit is contained in:
Jakub Sztandera 2020-06-30 00:31:05 +02:00
parent bc88c632d1
commit 21148033e2
No known key found for this signature in database
GPG Key ID: 9A9AF56F8B3879BA
2 changed files with 16 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
cmd/lotus-bench/import.go
View File

@ -232,21 +232,20 @@ func compStats(vals []float64) (float64, float64) {
type stats struct {
count float64
mean float64
dSqr float64
m2 float64
}
func (s *stats) AddPoint(value float64) {
s.count++
meanDiff := (value - s.mean) / s.count
newMean := s.mean + meanDiff
dSqrtInc := (value - newMean) * (value - s.mean)
s.dSqr += dSqrtInc
s.mean = newMean
// based on https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Welford's_online_algorithm
s.count += 1
delta := value - s.mean
s.mean += delta / s.count
delta2 := value - s.mean
s.m2 += delta * delta2
}
func (s *stats) variance() float64 {
return s.dSqr / (s.count - 1)
return s.m2 / (s.count - 1)
}
func (s1 *stats) Combine(s2 *stats) {
@ -268,14 +267,12 @@ func (s1 *stats) Combine(s2 *stats) {
}
newCount := s1.count + s2.count
newMean := s1.count*s1.mean + s2.count*s2.mean
newMean /= newCount
newVar := s1.count * (s1.variance() + (s1.mean-newMean)*(s1.mean-newMean))
newVar += s2.count * (s2.variance() + (s2.mean-newMean)*(s2.mean-newMean))
newVar /= newCount
delta := s2.mean - s1.mean
meanDelta := delta * s2.count / newCount
m2 := s1.m2 + s2.m2 + delta*meanDelta*s1.count
s1.count = newCount
s1.mean = newMean
s1.dSqr = newVar * (newCount - 1)
s1.mean += meanDelta
s1.m2 = m2
}
func tallyGasCharges(charges map[string]*stats, et types.ExecutionTrace) {

6
cmd/lotus-bench/stats_test.go
View File

@ -9,12 +9,12 @@ import (
func TestStats(t *testing.T) {
N := 16
ss := make([]*stats, N)
rng := rand.New(rand.NewSource(1))
for i := 0; i < N; i++ {
ss[i] = &stats{}
maxJ := rand.Intn(1000)
maxJ := rng.Intn(1000)
for j := 0; j < maxJ; j++ {
ss[i].AddPoint(rand.NormFloat64()*5 + 500)
ss[i].AddPoint(rand.NormFloat64()*5 + 1000)
ss[i].AddPoint(rng.NormFloat64()*5 + 500)
}
t.Logf("mean: %f, stddev: %f, count %f", ss[i].mean, math.Sqrt(ss[i].variance()), ss[i].count)
}