Visible to Intel only — GUID: GUID-BA245394-1393-48BB-B55E-88905056A531
Visible to Intel only — GUID: GUID-BA245394-1393-48BB-B55E-88905056A531
poisson_v
Generates Poisson distributed random values with varying mean.
Description
The poisson_v class object is used in the generate function to provide n Poisson distributed random numbers with distribution parameter , where .
The probability distribution is given by:
The cumulative distribution function is as follows:
Product and Performance Information |
---|
Performance varies by use, configuration, and other factors. Learn more at https://www.intel.com/PerformanceIndex. Notice revision #20201201 |
API
Syntax
namespace oneapi::mkl::rng { template<typename IntType = std::int32_t, typename Method = poisson_v_method::by_default> class poisson_v { public: using method_type = Method; using result_type = IntType; explicit poisson_v(std::vector<double> lambda); // deprecated since oneMKL 2023.0 explicit poisson_v(sycl::span<double> lambda); explicit poisson_v(const param_type& pt); std::vector<double> lambda() const; param_type param() const; void param(const param_type& pt); }; }
Devices supported: CPU and GPU
Include Files
oneapi/mkl/rng.hpp
Template Parameters
typename IntType = std::int32_t |
Type of the produced values. The specific values are as follows: std::int32_t std::uint32_t |
typename Method = oneapi::mkl::rng::poisson_v_method:: by_default |
Generation method. The specific values are as follows: oneapi::mkl::rng::poisson_v_method::gaussian_icdf_based See brief descriptions of the methods in Distributions Template Parameter Method. |
Input Parameters
Name |
Type |
Description |
---|---|---|
lambda |
sycl::span<double> |
Array of n distribution parameters λ. |
- explicit poisson_v(std::vector<double> lambda); is deprecated and will be removed in one of the next releases. Use explicit poisson_v(sycl::span<double> lambda); instead.
When passing a sycl::span that is constructed over a user’s memory to the constructor, users must manage the memory under sycl::span by themselves. They must not destroy the memory while data are processed.