Layer API

Neural network layer construction, evaluation, and parameters. More...

Data Structures
struct	mg_layer
	Layer of neurons in a neural network. More...

Functions
bool	mg_layer_init (mg_graph *g, mg_layer *l, size_t n_in, size_t n_out, bool non_linear)
	Initialize a layer.
bool	mg_layer_call (mg_graph *g, mg_layer *l, mg_value **x, mg_value **out)
	Evaluate a layer with the given input values.
void	mg_layer_free (mg_layer *l)
	Free memory owned by a layer.
size_t	mg_layer_param_count (const mg_layer *l)
	Get the number of trainable parameters in a layer.
void	mg_layer_params (const mg_layer *l, mg_value **out)
	Copy a layer's parameters into a flat array.

Detailed Description

Neural network layer construction, evaluation, and parameters.

Function Documentation

mg_layer_call()

bool mg_layer_call	(	mg_graph *	g,
		mg_layer *	l,
		mg_value **	x,
		mg_value **	out
	)

Evaluate a layer with the given input values.

On failure, the graph is left unchanged and every out entry written by this call is reset to NULL.

Parameters

g	Graph that owns any intermediate and output values.
l	Layer to evaluate.
x	Array of l->n_in input values.
out	Output array with at least l->n_out entries.

Returns

true on success, false on invalid input or allocation failure.

mg_layer_free()

void mg_layer_free

(

mg_layer *

l

)

Free memory owned by a layer.

Parameters

l	Layer to free.

mg_layer_init()

bool mg_layer_init	(	mg_graph *	g,
		mg_layer *	l,
		size_t	n_in,
		size_t	n_out,
		bool	non_linear
	)

Initialize a layer.

Parameters

g	Graph that owns the layer's values.
l	Layer to initialize.
n_in	Number of input values consumed by each neuron.
n_out	Number of output values produced by the layer.
non_linear	Whether each neuron should use a non-linear activation function.

Returns

true on success, false on allocation failure.

mg_layer_param_count()

size_t mg_layer_param_count

(

const mg_layer *

l

)

Get the number of trainable parameters in a layer.

Parameters

l	Layer to inspect.

Returns

Total number of parameters across all neurons.

mg_layer_params()

void mg_layer_params	(	const mg_layer *	l,
		mg_value **	out
	)

Copy a layer's parameters into a flat array.

Parameters

l	Layer to inspect.
out	Output array with at least mg_layer_param_count(l) entries.