cerc-io/lotus
15
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 27 Wiki Activity

Be even more precise - properly account for multi-win Poisson

Browse Source
This commit is contained in:
Peter Rabbitson 2021-06-14 23:41:43 +02:00
parent 4d1fccd290
commit ae2b089a5c
1 changed files with 17 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
cmd/lotus-storage-miner/info.go
View File

@ -165,18 +165,22 @@ func infoCmdAct(cctx *cli.Context) error {
fmt.Print("Below minimum power threshold, no blocks will be won")
} else {
winRatio := new(corebig.Rat).Mul(
new(corebig.Rat).SetFrac(
types.BigMul(pow.MinerPower.QualityAdjPower, types.NewInt(build.BlocksPerEpoch)).Int,
pow.TotalPower.QualityAdjPower.Int,
),
// decrease the rate ever-so-slightly to very roughly account for the multi-win poisson distribution
// FIXME - this is not a scientifically derived number... like at all
corebig.NewRat(99997, 100000),
winRatio := new(corebig.Rat).SetFrac(
types.BigMul(pow.MinerPower.QualityAdjPower, types.NewInt(build.BlocksPerEpoch)).Int,
pow.TotalPower.QualityAdjPower.Int,
)
if winRatioFloat, _ := winRatio.Float64(); winRatioFloat > 0 {
// if the corresponding poisson distribution isn't infinitely small then
// throw it into the mix as well, accounting for multi-wins
winRationWithPoissonFloat := -math.Expm1(-winRatioFloat)
winRationWithPoisson := new(corebig.Rat).SetFloat64(winRationWithPoissonFloat)
if winRationWithPoisson != nil {
winRatio = winRationWithPoisson
winRatioFloat = winRationWithPoissonFloat
}
weekly, _ := new(corebig.Rat).Mul(
winRatio,
new(corebig.Rat).SetInt64(7*builtin.EpochsInDay),
@ -194,8 +198,11 @@ func infoCmdAct(cctx *cli.Context) error {
(time.Second * time.Duration(avgDuration)).Truncate(time.Second).String(),
)
// Geometric distribution calculated as described in https://en.wikipedia.org/wiki/Geometric_distribution#Probability_Outcomes_Examples
// https://www.wolframalpha.com/input/?i=t+%3E+0%3B+p+%3E+0%3B+p+%3C+1%3B+%281-%28p%29%29%5Et%3D%281-%28c%2F100%29%29%3B+solve+t
// Geometric distribution of P(Y < k) calculated as described in https://en.wikipedia.org/wiki/Geometric_distribution#Probability_Outcomes_Examples
// https://www.wolframalpha.com/input/?i=t+%3E+0%3B+p+%3E+0%3B+p+%3C+1%3B+c+%3E+0%3B+c+%3C1%3B+1-%281-p%29%5E%28t%29%3Dc%3B+solve+t
// t == how many dice-rolls (epochs) before win
// p == winRate == ( minerPower / netPower )
// c == target probability of win ( 99.9% in this case )
fmt.Print("Projected block win with ")
color.Green(
"99.9%% probability every %s",