MayaFlux::Kinesis::Stochastic::Weights Class Reference

Rows-by-columns table of doubles with atomic cell access. More...

#include <Weights.hpp>

Collaboration diagram for MayaFlux::Kinesis::Stochastic::Weights:

Classes
class	Cell
	Proxy for one cell with atomic load and store. More...

Public Member Functions
void	apply_edges (std::span< const std::pair< size_t, size_t > > edges, double weight=1.0, bool symmetric=true, bool normalize=true) noexcept
	Write an adjacency edge list into this table.
void	assign (size_t rows, size_t cols, double initial=0.0)
	Replace dimensions and reset every cell.
size_t	cols () const noexcept
void	fill (double value) noexcept
double	get (size_t row, size_t col) const noexcept
	Atomic read of one cell.
void	normalize_row (size_t row) noexcept
	Scale a row so its cells sum to one.
Cell	operator() (size_t row, size_t col) noexcept
	Access one cell.
Weights &	operator= (std::initializer_list< std::initializer_list< double > > values)
	Assign every cell from nested initializer lists.
std::span< const double >	row (size_t row) const noexcept
	Non-atomic view of one row.
size_t	rows () const noexcept
void	scale_row (size_t row, double factor) noexcept
void	uniform_excluding_self () noexcept
	Every cell uniform except the diagonal, each row normalized to sum to one.
size_t	weighted_pick (size_t row_index, Stochastic &generator) const
	Weighted random draw over one row of this table.
	Weights ()=default
	Weights (size_t rows, size_t cols, double initial=0.0)

Static Public Member Functions
static size_t	weighted_pick (std::span< const double > weights, Stochastic &generator)
	Weighted random draw over a row of nonnegative weights.

Private Attributes
size_t	m_cols {}
std::vector< double >	m_data
size_t	m_rows {}

Detailed Description

Rows-by-columns table of doubles with atomic cell access.

A plain numeric table with no knowledge of what its rows and columns mean. Transition probabilities, source-to-output weights, and neighbour strengths are the same structure read by different code.

Cells are read and written with relaxed atomics, so any thread can write a cell while another reads it. Row helpers and row views are not atomic and belong on the thread that consumes the table. Copying is not synchronised. Dimensions change only through assign(), which must not run concurrently with any other access.

weighted_pick() turns a row into a single stochastic choice, drawing from a Stochastic instance the caller supplies and owns (one instance per thread, matching Stochastic's own contract).

Usage

Weights transitions(5, 5);
transitions.normalize_row(0);
 
Stochastic draw(Algorithm::UNIFORM);
size_t next = transitions.weighted_pick(0, draw);

Definition at line 36 of file Weights.hpp.

---

The documentation for this class was generated from the following file:

• Weights.hpp