package main import ( "fmt" "git.nunosempere.com/NunoSempere/probppl/choose" "math" rand "math/rand/v2" "sync" ) type src = *rand.Rand type IntProb struct { N int64 p float64 } type IntProbs = []IntProb type IntProbsWeights struct { IntProbs IntProbs w int64 } func generatePeopleKnownDistribution(r src) IntProbs { var probabilities IntProbs sum := 0.0 for i := 16; i <= 2048; i *= 2 { p := r.Float64() probabilities = append(probabilities, IntProb{N: int64(i), p: p}) sum += p } for i := range probabilities { probabilities[i].p = probabilities[i].p / sum } return probabilities } func chooseWrapper(n int64, k int64) int64 { if n < k { return 0 } else { return choose.Choose(n, k) } } func getProbabilityOfKBirthdayMatchesGivenNPeopleKnown(n int64, k int64) float64 { return float64(chooseWrapper(n, k)) * math.Pow(1.0/365.0, float64(k)) * math.Pow(1.0-(1.0/365.0), float64(n-k)) } func getMatchesDrawGivenNPeopleKnown(n int64, r src) int64 { p0 := getProbabilityOfKBirthdayMatchesGivenNPeopleKnown(n, 0) p1 := getProbabilityOfKBirthdayMatchesGivenNPeopleKnown(n, 1) p2 := getProbabilityOfKBirthdayMatchesGivenNPeopleKnown(n, 2) p := r.Float64() if p < p0 { return 0 } else if p < (p0 + p1) { return 1 } else if p < (p0 + p1 + p2) { return 2 } else { return 3 // stands for 'greater than 3' } } /* Draw 148 times How many people do you know that were born in the same day of the year as you? 0: 46.6% | 69 1: 31.1% | 46 2: 12.8% | 19 ≥3: 9.5% | 14 */ func drawFromDistributionWithReplacement(d IntProbs, r src) int64 { pp := r.Float64() sum := 0.0 for i := range d { sum += d[i].p if pp <= sum { return int64(d[i].N) // this introduces some non-determinism, as order of maps in go isn't guaranteed } } fmt.Printf("%f, %f\n", sum, pp) fmt.Println(d) panic("Probabilities should sum up to 1") } func aboutEq(a int64, b int64) bool { h := int64(3) return ((-h) <= (a - b)) && ((a - b) <= h) } func draw148PplFromDistributionAndCheck(d IntProbs, r src, show bool) int64 { count := make(map[int64]int64) count[0] = 0 count[1] = 0 count[2] = 0 count[3] = 0 for i := 0; i < 148; i++ { person_i_ppl_known := drawFromDistributionWithReplacement(d, r) person_i_num_birthday_matches := getMatchesDrawGivenNPeopleKnown(person_i_ppl_known, r) count[person_i_num_birthday_matches]++ } // if (count[0] == 69) && (count[1] == 46) && (count[2] == 19) && (count[3] == 14) { if show { // fmt.Println(count) } if aboutEq(count[0], 69) && aboutEq(count[1], 46) && aboutEq(count[2], 19) && aboutEq(count[3], 14) { return 1 } else { return 0 } } func getUnnormalizedBayesianUpdateForDistribution(d IntProbs, r src) int64 { var sum int64 = 0 n := 30_000 for i := 0; i < n; i++ { /* if i%1000 == 0 { fmt.Println(i) } */ draw_result := draw148PplFromDistributionAndCheck(d, r, i == 0) // fmt.Println(draw_result) sum += draw_result } return sum // float64(sum) / float64(n) } func main() { n_dists := 30_000 var dists = make([]IntProbsWeights, n_dists) // Prepare for concurrency num_threads := 32 var wg sync.WaitGroup wg.Add(num_threads) for i := range num_threads { go func() { defer wg.Done() var r = rand.New(rand.NewPCG(uint64(i), uint64(i+1))) for j := i * (n_dists / num_threads); j < (i+1)*(n_dists/num_threads); j++ { people_known_distribution := generatePeopleKnownDistribution(r) result := getUnnormalizedBayesianUpdateForDistribution(people_known_distribution, r) /* if i%10 == 0 { fmt.Printf("%d/%d\n", i, n_dists) } */ if result > 0 { dists[j] = IntProbsWeights{IntProbs: people_known_distribution, w: result} } } }() } wg.Wait() // Now calculate the posterior sum_weights := int64(0) for _, dist := range dists { sum_weights += dist.w } for i := int64(16); i <= 2048; i *= 2 { p := 0.0 for _, dist := range dists { for _, int_prob := range dist.IntProbs { if int_prob.N == i { p += float64(dist.w) * int_prob.p } } } p = p / float64(sum_weights) fmt.Printf("%d: %f\n", i, p) } }