GammaPoisson.h

Go to the documentation of this file.

 
#ifndef EIGENRAND_DISTS_GAMMAPOISSON_H
#define EIGENRAND_DISTS_GAMMAPOISSON_H
 
#include <memory>
#include <iterator>
#include <limits>
 
namespace Eigen
{
    namespace Rand
    {
        template<typename _Scalar>
        class NegativeBinomialGen : public GenBase<NegativeBinomialGen<_Scalar>, _Scalar>
        {
            static_assert(std::is_same<_Scalar, int32_t>::value, "negativeBinomial needs integral types."); 
            UniformRealGen<float> ur;
            GammaGen<float> gamma;
        public:
            using Scalar = _Scalar;
 
            NegativeBinomialGen(_Scalar _trials = 1, double _p = 0.5)
                : gamma{ (float)_trials, (float)((1 - _p) / _p) }
 
            {
            }
 
            template<typename Rng>
            EIGEN_STRONG_INLINE const _Scalar operator() (Rng&& rng)
            {
                using namespace Eigen::internal;
                float v = gamma(rng);
                return PoissonGen<_Scalar>{v}(rng);
            }
 
            template<typename Packet, typename Rng>
            EIGEN_STRONG_INLINE const Packet packetOp(Rng&& rng)
            {
                using namespace Eigen::internal;
                using PacketType = decltype(reinterpret_to_float(std::declval<Packet>()));
 
                auto mean = gamma.template packetOp<PacketType>(rng);
                auto res = pset1<Packet>(0);
                PacketType val = pset1<PacketType>(1), pne_mean = pexp(pnegate(mean));
                if (pmovemask(pcmplt(pset1<PacketType>(12), mean)) == 0)
                {
                    for (int _i = 0; ; ++_i)
                    {
                        EIGENRAND_CHECK_INFINITY_LOOP();
                        val = pmul(val, ur.template packetOp<PacketType>(rng));
                        auto c = reinterpret_to_int(pcmplt(pne_mean, val));
                        if (pmovemask(c) == 0) break;
                        res = padd(res, pnegate(c));
                    }
                    return res;
                }
                else
                {
                    auto& cm = Rand::detail::CompressMask<sizeof(Packet)>::get_inst();
                    const PacketType ppi = pset1<PacketType>(constant::pi),
                        psqrt_tmean = psqrt(pmul(pset1<PacketType>(2), mean)),
                        plog_mean = plog(mean),
                        pg1 = psub(pmul(mean, plog_mean), plgamma_approx(padd(mean, pset1<PacketType>(1))));
                    for (int _i = 0; ; ++_i)
                    {
                        EIGENRAND_CHECK_INFINITY_LOOP();
                        PacketType fres, yx, psin, pcos;
                        psincos(pmul(ppi, ur.template packetOp<PacketType>(rng)), psin, pcos);
                        yx = pdiv(psin, pcos);
                        fres = ptruncate(padd(pmul(psqrt_tmean, yx), mean));
 
                        auto p1 = pmul(padd(pmul(yx, yx), pset1<PacketType>(1)), pset1<PacketType>(0.9));
                        auto p2 = pexp(psub(psub(pmul(fres, plog_mean), plgamma_approx(padd(fres, pset1<PacketType>(1)))), pg1));
 
                        auto c1 = pcmple(pset1<PacketType>(0), fres);
                        auto c2 = pcmple(ur.template packetOp<PacketType>(rng), pmul(p1, p2));
 
                        auto cands = fres;
                        bool full = false;
                        gamma.cache_rest_cnt = cm.compress_append(cands, pand(c1, c2),
                            gamma.template get<PacketType>(), gamma.cache_rest_cnt, full);
                        if (full) return pcast<PacketType, Packet>(cands);
                    }
                }
            }
        };
 
        template<typename Derived, typename Urng>
        using NegativeBinomialType = CwiseNullaryOp<internal::scalar_rng_adaptor<NegativeBinomialGen<typename Derived::Scalar>, typename Derived::Scalar, Urng, true>, const Derived>;
 
        template<typename Derived, typename Urng>
        inline const NegativeBinomialType<Derived, Urng>
            negativeBinomial(Index rows, Index cols, Urng&& urng, typename Derived::Scalar trials = 1, double p = 0.5)
        {
            return {
                rows, cols, { std::forward<Urng>(urng), NegativeBinomialGen<typename Derived::Scalar>{trials, p} }
            };
        }
 
        template<typename Derived, typename Urng>
        inline const NegativeBinomialType<Derived, Urng>
            negativeBinomialLike(Derived& o, Urng&& urng, typename Derived::Scalar trials = 1, double p = 0.5)
        {
            return {
                o.rows(), o.cols(), { std::forward<Urng>(urng), NegativeBinomialGen<typename Derived::Scalar>{trials, p} }
            };
        }
    }
}
#endif