construct_program()

template<typename... Ts>

program_rep construct_program	(	gr_pair< Ts... > &	gr,
		program_info &	info,
		std::optional< std::string >	file_ = std::nullopt
	)

Translates a graph_ir into a sample-based graphical model library built on ccyan primitives.

Given a directed graphical model represented as a graph_ir, this function does multiple things:

• defines observed and unobserved structs that hold values of observed and unobserved random variables and their associated log probabilities
• defines a function to compute the joint log probability of the model
• defines functions that are equivalent to the following interpretations of the graphical model: (a) running it forward conditioned on observed data; (b) evaluating the joint log probability of the model conditioned on both observed and unobserved random variables; (c) predicting values of observed random variables conditioned on unobserved random variables.
• creates associated build utilities (CMakeLists.txt files, blank canvas main file)

• Todo:

  more functionality coming – a work in progress

These functions are implemented using ccyan primitives. ccyan is a C99 library that depends only on libm and stdint. It does not dynamically allocate memory, making it safe for many low-resource or real-time applications.

Template Parameters

Ts	The distribution types in the graph_ir

Parameters

gr	a graph_ir and a record pair
info	information about the generated program

Todo:

more information – a work in progress

Parameters

file_

if not null, the directory in which to save the generated code relative to executable directory (e.g., "out/")

Returns

program_rep