Neuron API

Fully connected neuron construction, evaluation, and parameters. More...

Data Structures
struct	mg_neuron
	Fully connected neuron in a neural network. More...

Functions
bool	mg_neuron_init (mg_graph *g, mg_neuron *n, size_t nin, bool non_linear)
	Initialize a neuron.
void	mg_neuron_free (mg_neuron *n)
	Free memory owned by a neuron.
mg_value *	mg_neuron_call (mg_graph *g, mg_neuron *n, mg_value **x)
	Evaluate a neuron with the given input values.
size_t	mg_neuron_param_count (const mg_neuron *n)
	Get the number of trainable parameters in a neuron.
void	mg_neuron_params (const mg_neuron *n, mg_value **out)
	Copy a neuron's parameters into a flat array.

Detailed Description

Fully connected neuron construction, evaluation, and parameters.

Function Documentation

mg_neuron_call()

mg_value * mg_neuron_call	(	mg_graph *	g,
		mg_neuron *	n,
		mg_value **	x
	)

Evaluate a neuron with the given input values.

Parameters

g	Graph that owns any intermediate and output values.
n	Neuron to evaluate.
x	Array of n->n_in input values.

Returns

The output value, or NULL on invalid input or allocation failure.

mg_neuron_free()

void mg_neuron_free

(

mg_neuron *

n

)

Free memory owned by a neuron.

Parameters

n	Neuron to free.

mg_neuron_init()

bool mg_neuron_init	(	mg_graph *	g,
		mg_neuron *	n,
		size_t	nin,
		bool	non_linear
	)

Initialize a neuron.

Parameters

g	Graph that owns the neuron's values.
n	Neuron to initialize.
nin	Number of input values.
non_linear	Whether to apply the non-linear activation function.

Returns

true on success, false on allocation failure.

mg_neuron_param_count()

size_t mg_neuron_param_count

(

const mg_neuron *

n

)

Get the number of trainable parameters in a neuron.

Parameters

n	Neuron to inspect.

Returns

Number of weight values plus the bias value.

mg_neuron_params()

void mg_neuron_params	(	const mg_neuron *	n,
		mg_value **	out
	)

Copy a neuron's parameters into a flat array.

Parameters

n	Neuron to inspect.
out	Output array with at least mg_neuron_param_count(n) entries.