An implementation of a multi-layer perceptron neural network. More...

#include "neural.h"
#include "neural_layer_connected.h"
#include "neural_layer_dropout.h"
#include "neural_layer_noise.h"
#include "neural_layer_recurrent.h"
#include "neural_layer_softmax.h"
#include "utils.h"

Include dependency graph for neural.c:

Functions
void	neural_init (struct Net *net)
	Initialises an empty neural network.

void	neural_create (struct Net net, struct ArgsLayer arg)
	Initialises and creates a new neural network from a parameter list.

void	neural_insert (struct Net net, struct Layer l, const int pos)
	Inserts a layer into a neural network.

void	neural_remove (struct Net *net, const int pos)
	Removes a layer from a neural network.

void	neural_push (struct Net net, struct Layer l)
	Inserts a layer at the head of a neural network.

void	neural_pop (struct Net *net)
	Removes the layer at the head of a neural network.

void	neural_copy (struct Net dest, const struct Net src)
	Copies a neural network.

void	neural_free (struct Net *net)
	Frees a neural network.

void	neural_rand (const struct Net *net)
	Randomises the layers within a neural network.

bool	neural_mutate (const struct Net *net)
	Mutates a neural network.

void	neural_resize (const struct Net *net)
	Resizes neural network layers as necessary.

void	neural_propagate (struct Net net, const double input, const bool train)
	Forward propagates a neural network.

void	neural_learn (const struct Net net, const double truth, const double *input)
	Performs a gradient descent update on a neural network.

double	neural_output (const struct Net *net, const int IDX)
	Returns the output of a specified neuron in the output layer of a neural network.

double *	neural_outputs (const struct Net *net)
	Returns the outputs from the output layer of a neural network.

void	neural_print (const struct Net *net, const bool print_weights)
	Prints a neural network.

char *	neural_json_export (const struct Net *net, const bool return_weights)
	Returns a json formatted string representation of a neural network.

void	neural_json_import (struct Net net, const struct ArgsLayer arg, const cJSON *json)
	Creates a neural network from a cJSON object.

double	neural_size (const struct Net *net)
	Returns the total number of non-zero weights in a neural network.

size_t	neural_save (const struct Net net, FILE fp)
	Writes a neural network to a file.

size_t	neural_load (struct Net net, FILE fp)
	Reads a neural network from a file.

Detailed Description

An implementation of a multi-layer perceptron neural network.

Author: Richard Preen rpree.nosp@m.n@gm.nosp@m.ail.c.nosp@m.om

Copyright: The Authors.

Date: 2012–2024.

Definition in file neural.c.

Function Documentation

◆ neural_copy()

void neural_copy	(	struct Net *	dest,
		const struct Net *	src
	)

Copies a neural network.

Parameters

[in]	dest	The destination neural network.
[in]	src	The source neural network.

Definition at line 208 of file neural.c.

References Llist::layer, layer_copy(), neural_init(), neural_push(), Llist::prev, and Net::tail.

Referenced by act_neural_copy(), cond_neural_copy(), pred_neural_copy(), and rule_neural_cond_copy().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_create()

void neural_create	(	struct Net *	net,
		struct ArgsLayer *	arg
	)

Initialises and creates a new neural network from a parameter list.

Parameters

[in]	net	The neural network to initialise.
[in]	arg	List of layer parameters defining the initial network.

Definition at line 54 of file neural.c.

References AVGPOOL, ArgsLayer::channels, DROPOUT, ArgsLayer::height, layer_init(), MAXPOOL, ArgsLayer::n_init, Net::n_inputs, ArgsLayer::n_inputs, Net::n_layers, Net::n_outputs, Layer::n_outputs, neural_init(), neural_push(), ArgsLayer::next, NOISE, Layer::out_c, Layer::out_h, Layer::out_w, SOFTMAX, ArgsLayer::type, UPSAMPLE, and ArgsLayer::width.

Referenced by act_neural_init(), cond_neural_init(), layer_args_json_export(), pred_neural_init(), and rule_neural_cond_init().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_free()

void neural_free ( struct Net * net )

Frees a neural network.

Parameters

[in] net The neural network to free.

Definition at line 224 of file neural.c.

References Llist::layer, layer_free(), Net::n_layers, Llist::prev, and Net::tail.

Referenced by act_neural_free(), cond_neural_free(), layer_args_json_export(), pred_neural_free(), and rule_neural_cond_free().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_init()

void neural_init ( struct Net * net )

Initialises an empty neural network.

Parameters

[in] net The neural network to initialise.

Definition at line 37 of file neural.c.

References Net::head, Net::n_inputs, Net::n_layers, Net::n_outputs, Net::output, Net::tail, and Net::train.

Referenced by layer_args_json_export(), neural_copy(), neural_create(), and neural_load().

Here is the caller graph for this function:

◆ neural_insert()

void neural_insert	(	struct Net *	net,
		struct Layer *	l,
		const int	pos
	)

Inserts a layer into a neural network.

Parameters

[in]	net	The neural network receiving the layer.
[in]	l	The layer to insert.
[in]	pos	The position in the network to insert the layer.

Definition at line 95 of file neural.c.

References Net::head, Llist::layer, Net::n_inputs, Layer::n_inputs, Net::n_layers, Net::n_outputs, Layer::n_outputs, Llist::next, Net::output, Layer::output, Llist::prev, and Net::tail.

Referenced by neural_push(), and pred_neural_expand().

Here is the caller graph for this function:

◆ neural_json_export()

char * neural_json_export	(	const struct Net *	net,
		const bool	return_weights
	)

Returns a json formatted string representation of a neural network.

Parameters

[in]	net	The neural network to return.
[in]	return_weights	Whether to return the weights in each layer.

Returns: String encoded in json format.

Definition at line 407 of file neural.c.

References Llist::layer, layer_json_export(), Llist::prev, and Net::tail.

Referenced by act_neural_json_export(), cond_neural_json_export(), neural_print(), pred_neural_json_export(), and rule_neural_cond_json_export().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_json_import()

void neural_json_import	(	struct Net *	net,
		const struct ArgsLayer *	arg,
		const cJSON *	json
	)

Creates a neural network from a cJSON object.

Parameters

[in,out]	net	The neural network to return.
[in]	arg	List of layer parameters defining the initial network.
[in]	json	cJSON object.

Definition at line 434 of file neural.c.

Referenced by act_neural_json_import(), cond_neural_json_import(), pred_neural_json_import(), and rule_neural_cond_json_import().

Here is the caller graph for this function:

◆ neural_learn()

void neural_learn	(	const struct Net *	net,
		const double *	truth,
		const double *	input
	)

Performs a gradient descent update on a neural network.

Parameters

[in]	net	The neural network to be updated.
[in]	truth	The desired network output.
[in]	input	The input state.

Definition at line 328 of file neural.c.

References Layer::delta, Net::head, Layer::i, Llist::layer, layer_backward(), layer_update(), Layer::n_outputs, Llist::next, Layer::output, Llist::prev, and Net::tail.

Referenced by pred_neural_update().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_load()

size_t neural_load	(	struct Net *	net,
		FILE *	fp
	)

Reads a neural network from a file.

Parameters

[in]	net	The neural network to load.
[in]	fp	Pointer to the file to be read.

Returns: The number of elements read.

Definition at line 499 of file neural.c.

References layer_load(), neural_init(), and neural_push().

Referenced by act_neural_load(), cond_neural_load(), pred_neural_load(), and rule_neural_cond_load().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_mutate()

bool neural_mutate ( const struct Net * net )

Mutates a neural network.

Parameters

[in] net The neural network to mutate.

Returns: Whether any alterations were made.

Definition at line 257 of file neural.c.

References Llist::layer, layer_mutate(), layer_resize(), Layer::n_outputs, Llist::prev, and Net::tail.

Referenced by act_neural_mutate(), cond_neural_mutate(), pred_neural_mutate(), and rule_neural_cond_mutate().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_output()

double neural_output	(	const struct Net *	net,
		const int	IDX
	)

Returns the output of a specified neuron in the output layer of a neural network.

Parameters

[in]	net	The neural network to output.
[in]	IDX	Which neuron in the output layer to return.

Returns: The output of the specified neuron.

Definition at line 369 of file neural.c.

References Net::head, Llist::layer, layer_output(), and Net::n_outputs.

Referenced by cond_neural_match(), pred_neural_compute(), rule_neural_act_compute(), and rule_neural_cond_match().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_outputs()

double * neural_outputs ( const struct Net * net )

Returns the outputs from the output layer of a neural network.

Parameters

[in] net The neural network to output.

Returns: The neural network outputs.

Definition at line 384 of file neural.c.

References Net::head, Llist::layer, and layer_output().

Referenced by act_neural_compute().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_pop()

void neural_pop ( struct Net * net )

Removes the layer at the head of a neural network.

Parameters

[in] net The neural network receiving the layer.

Definition at line 197 of file neural.c.

References Net::n_layers, and neural_remove().

Referenced by pred_neural_ae_to_classifier().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_print()

void neural_print	(	const struct Net *	net,
		const bool	print_weights
	)

Prints a neural network.

Parameters

[in]	net	The neural network to print.
[in]	print_weights	Whether to print the weights in each layer.

Definition at line 395 of file neural.c.

References neural_json_export().

Here is the call graph for this function:

◆ neural_propagate()

void neural_propagate	(	struct Net *	net,
		const double *	input,
		const bool	train
	)

Forward propagates a neural network.

Parameters

[in]	net	Neural network to propagate.
[in]	input	Input state.
[in]	train	Whether the network is in training mode.

Definition at line 310 of file neural.c.

References Llist::layer, layer_forward(), layer_output(), Llist::prev, Net::tail, and Net::train.

Referenced by act_neural_compute(), cond_neural_match(), pred_neural_compute(), and rule_neural_cond_match().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_push()

void neural_push	(	struct Net *	net,
		struct Layer *	l
	)

Inserts a layer at the head of a neural network.

Parameters

[in]	net	The neural network receiving the layer.
[in]	l	The layer to insert.

Definition at line 187 of file neural.c.

References Net::n_layers, and neural_insert().

Referenced by neural_copy(), neural_create(), neural_load(), and pred_neural_ae_to_classifier().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_rand()

void neural_rand ( const struct Net * net )

Randomises the layers within a neural network.

Parameters

[in] net The neural network to randomise.

Definition at line 242 of file neural.c.

References Llist::layer, layer_rand(), Llist::prev, and Net::tail.

Referenced by act_neural_cover(), cond_neural_cover(), and rule_neural_act_cover().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_remove()

void neural_remove	(	struct Net *	net,
		const int	pos
	)

Removes a layer from a neural network.

Parameters

[in]	net	The neural network removing the layer.
[in]	pos	The position of the layer in the network to be removed.

Definition at line 140 of file neural.c.

References Net::head, Llist::layer, layer_free(), Net::n_layers, Net::n_outputs, Layer::n_outputs, Llist::next, Net::output, Layer::output, Llist::prev, and Net::tail.

Referenced by neural_pop().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_resize()

void neural_resize ( const struct Net * net )

Resizes neural network layers as necessary.

Parameters

[in] net The neural network to resize.

Definition at line 290 of file neural.c.

References Llist::layer, layer_resize(), Layer::n_inputs, Llist::prev, and Net::tail.

Referenced by pred_neural_expand().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_save()

size_t neural_save	(	const struct Net *	net,
		FILE *	fp
	)

Writes a neural network to a file.

Parameters

[in]	net	The neural network to save.
[in]	fp	Pointer to the file to be written.

Returns: The number of elements written.

Definition at line 478 of file neural.c.

References Llist::layer, layer_save(), Net::n_inputs, Net::n_layers, Net::n_outputs, Llist::prev, and Net::tail.

Referenced by act_neural_save(), cond_neural_save(), pred_neural_save(), and rule_neural_cond_save().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ neural_size()

double neural_size ( const struct Net * net )

Returns the total number of non-zero weights in a neural network.

Parameters

[in] net A neural network.

Returns: The calculated network size.

Definition at line 450 of file neural.c.

References CONNECTED, CONVOLUTIONAL, Llist::layer, LSTM, Layer::n_active, Llist::prev, RECURRENT, Layer::size, Net::tail, and Layer::type.

Referenced by cond_neural_size(), pred_neural_size(), and rule_neural_cond_size().

Here is the caller graph for this function:

Functions

Detailed Description

Function Documentation

◆ neural_copy()

◆ neural_create()

◆ neural_free()

◆ neural_init()

◆ neural_insert()

◆ neural_json_export()

◆ neural_json_import()

◆ neural_learn()

◆ neural_load()

◆ neural_mutate()

◆ neural_output()

◆ neural_outputs()

◆ neural_pop()

◆ neural_print()

◆ neural_propagate()

◆ neural_push()

◆ neural_rand()

◆ neural_remove()

◆ neural_resize()

◆ neural_save()

◆ neural_size()