Open
Description
Hi,
We spotted an apparent bias in the inverse cumulative poisson distribution function,
looking deeper, we found boost is generating different results when compared with scipy and statslib (https://github.com/kthohr/stats)
The value of mean is not important, but for p < 0.5, the result is offset by -1
ie:
double mean = 2719.13;
double p = 0.49;
double r_boost = boost::math::quantile(boost::math::poisson_distribution<>(mean), p);
double r_stats = stats::qpois(p,mean);
r_boost --> 2717
r_stats --> 2718
In Scipy:
import scipy.stats
mean = 2719.13
r = 0.49
print(scipy.stats.poisson.ppf(r, mean))
printout --> 2718.0
Code:
// Alternative statistics library: statslib
// https://github.com/kthohr/stats
#include <boost/math/distributions/poisson.hpp>
#include <stats.hpp>
#include <iostream>
double mean = 2719.13;
void test_range( double a, double b )
{
for (double p = a; p < b+0.0001; p += 0.01)
{
double r_boost = boost::math::quantile(boost::math::poisson_distribution<>(mean), p);
double r_stats = stats::qpois(p,mean);
double diff = r_stats - r_boost;
std::cout << "Mean:" << mean << " p:" << p << " Boost=" << r_boost << " statslib=" << r_stats << " difference=" << diff << (diff != 0 ? " DIFFERENT" : " same") << std::endl;
}
std::cout << std::endl;
}
int main()
{
test_range(0.01, 0.04);
test_range(0.47, 0.52);
test_range(0.96, 0.99);
return 0;
}
Output:
Mean:2719.13 p:0.01 Boost=2598 statslib=2599 difference=1 DIFFERENT
Mean:2719.13 p:0.02 Boost=2612 statslib=2613 difference=1 DIFFERENT
Mean:2719.13 p:0.03 Boost=2620 statslib=2621 difference=1 DIFFERENT
Mean:2719.13 p:0.04 Boost=2627 statslib=2628 difference=1 DIFFERENT
Mean:2719.13 p:0.47 Boost=2714 statslib=2715 difference=1 DIFFERENT
Mean:2719.13 p:0.48 Boost=2715 statslib=2716 difference=1 DIFFERENT
Mean:2719.13 p:0.49 Boost=2717 statslib=2718 difference=1 DIFFERENT
Mean:2719.13 p:0.5 Boost=2719 statslib=2719 difference=0 same
Mean:2719.13 p:0.51 Boost=2720 statslib=2720 difference=0 same
Mean:2719.13 p:0.52 Boost=2722 statslib=2722 difference=0 same
Mean:2719.13 p:0.96 Boost=2811 statslib=2811 difference=0 same
Mean:2719.13 p:0.97 Boost=2818 statslib=2818 difference=0 same
Mean:2719.13 p:0.98 Boost=2827 statslib=2827 difference=0 same
Mean:2719.13 p:0.99 Boost=2841 statslib=2841 difference=0 same
Metadata
Metadata
Assignees
Labels
No labels