Multi-layer perceptron neural network prediction functions. More...

#include "pred_neural.h"
#include "neural_activations.h"
#include "neural_layer_avgpool.h"
#include "neural_layer_connected.h"
#include "neural_layer_convolutional.h"
#include "neural_layer_dropout.h"
#include "neural_layer_lstm.h"
#include "neural_layer_maxpool.h"
#include "neural_layer_noise.h"
#include "neural_layer_recurrent.h"
#include "neural_layer_softmax.h"
#include "neural_layer_upsample.h"
#include "utils.h"

Include dependency graph for pred_neural.c:

Functions
void	pred_neural_init (const struct XCSF xcsf, struct Cl c)
	Creates and initialises a neural network prediction. More...

void	pred_neural_free (const struct XCSF xcsf, const struct Cl c)
	Frees the memory used by a neural network prediction. More...

void	pred_neural_copy (const struct XCSF xcsf, struct Cl dest, const struct Cl *src)
	Copies a neural network prediction from one classifier to another. More...

void	pred_neural_update (const struct XCSF xcsf, const struct Cl c, const double x, const double y)
	Backward propagates and updates a neural network prediction. More...

void	pred_neural_compute (const struct XCSF xcsf, const struct Cl c, const double *x)
	Forward propagates a neural network prediction with a provided input. More...

void	pred_neural_print (const struct XCSF xcsf, const struct Cl c)
	Prints a neural network prediction. More...

bool	pred_neural_crossover (const struct XCSF xcsf, const struct Cl c1, const struct Cl *c2)
	Dummy function since neural predictions do not perform crossover. More...

bool	pred_neural_mutate (const struct XCSF xcsf, const struct Cl c)
	Mutates a neural network prediction. More...

double	pred_neural_size (const struct XCSF xcsf, const struct Cl c)
	Returns the size of a neural network prediction. More...

size_t	pred_neural_save (const struct XCSF xcsf, const struct Cl c, FILE *fp)
	Writes a neural network prediction to a file. More...

size_t	pred_neural_load (const struct XCSF xcsf, struct Cl c, FILE *fp)
	Reads a neural network prediction from a file. More...

double	pred_neural_eta (const struct XCSF xcsf, const struct Cl c, const int layer)
	Returns the gradient descent rate of a neural prediction layer. More...

int	pred_neural_neurons (const struct XCSF xcsf, const struct Cl c, const int layer)
	Returns the number of neurons in a neural prediction layer. More...

int	pred_neural_connections (const struct XCSF xcsf, const struct Cl c, const int layer)
	Returns the number of active connections in a neural prediction layer. More...

int	pred_neural_layers (const struct XCSF xcsf, const struct Cl c)
	Returns the number of layers within a neural network prediction. More...

void	pred_neural_expand (const struct XCSF xcsf, const struct Cl c)
	Creates and inserts a hidden layer before the prediction output layer. More...

void	pred_neural_ae_to_classifier (const struct XCSF xcsf, const struct Cl c, const int n_del)
	Removes prediction (decoder) layers and inserts softmax output layer. More...

char *	pred_neural_json_export (const struct XCSF xcsf, const struct Cl c)
	Returns a json formatted string representation of a neural prediction. More...

void	pred_neural_json_import (const struct XCSF xcsf, struct Cl c, const cJSON *json)
	Creates a neural prediction from a cJSON object. More...

void	pred_neural_param_defaults (struct XCSF *xcsf)
	Initialises default neural prediction parameters. More...

char *	pred_neural_param_json_import (struct XCSF xcsf, cJSON json)
	Sets the neural network parameters from a cJSON object. More...

Detailed Description

Multi-layer perceptron neural network prediction functions.

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

Copyright: The Authors.

Date: 2016–2022.

Definition in file pred_neural.c.

Function Documentation

◆ pred_neural_ae_to_classifier()

void pred_neural_ae_to_classifier	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const int	n_del
	)

Removes prediction (decoder) layers and inserts softmax output layer.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier to convert from autoencoding to classification.
[in]	n_del	Number of (decoder) layers to remove from the network.

Definition at line 349 of file pred_neural.c.

References CONNECTED, ArgsLayer::decay, ArgsLayer::eta, ArgsLayer::eta_min, ArgsLayer::evolve_connect, ArgsLayer::evolve_eta, ArgsLayer::evolve_weights, Layer::i, layer_args_init(), layer_init(), LINEAR, ArgsLayer::momentum, Net::n_layers, Net::n_outputs, PredNeural::net, neural_pop(), neural_push(), Cl::pred, ArgsLayer::sgd_weights, and SOFTMAX.

Referenced by xcsf_ae_to_classifier().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred_neural_compute()

void pred_neural_compute	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const double *	x
	)

Forward propagates a neural network prediction with a provided input.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier calculating the prediction.
[in]	x	The input state.

Definition at line 106 of file pred_neural.c.

References PredNeural::net, neural_output(), neural_propagate(), Cl::pred, and Cl::prediction.

Here is the call graph for this function:

◆ pred_neural_connections()

int pred_neural_connections	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const int	layer
	)

Returns the number of active connections in a neural prediction layer.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	Classifier maintaining a neural network prediction.
[in]	layer	Position of a layer in the network.

Returns: The current number of active (non-zero) connections in a layer.

Definition at line 268 of file pred_neural.c.

References Llist::layer, Layer::n_active, PredNeural::net, Cl::pred, Llist::prev, and Net::tail.

Referenced by clset_mean_pred_connections().

Here is the caller graph for this function:

◆ pred_neural_copy()

void pred_neural_copy	(	const struct XCSF *	xcsf,
		struct Cl *	dest,
		const struct Cl *	src
	)

Copies a neural network prediction from one classifier to another.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	dest	The destination classifier.
[in]	src	The source classifier.

Definition at line 73 of file pred_neural.c.

References PredNeural::net, neural_copy(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_crossover()

bool pred_neural_crossover	(	const struct XCSF *	xcsf,
		const struct Cl *	c1,
		const struct Cl *	c2
	)

Dummy function since neural predictions do not perform crossover.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c1	The first classifier whose prediction is being crossed.
[in]	c2	The second classifier whose prediction is being crossed.

Returns: False.

Definition at line 137 of file pred_neural.c.

◆ pred_neural_eta()

double pred_neural_eta	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const int	layer
	)

Returns the gradient descent rate of a neural prediction layer.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	Classifier maintaining a neural network prediction.
[in]	layer	Position of a layer in the network.

Returns: The current gradient descent rate of a layer.

Definition at line 216 of file pred_neural.c.

References Layer::eta, Llist::layer, PredNeural::net, Cl::pred, Llist::prev, and Net::tail.

Referenced by clset_mean_pred_eta().

Here is the caller graph for this function:

◆ pred_neural_expand()

void pred_neural_expand	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Creates and inserts a hidden layer before the prediction output layer.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose neural prediction is to be expanded.

Definition at line 306 of file pred_neural.c.

References CONNECTED, ArgsLayer::decay, ArgsLayer::eta, ArgsLayer::eta_min, ArgsLayer::evolve_connect, ArgsLayer::evolve_eta, ArgsLayer::evolve_functions, ArgsLayer::evolve_weights, ArgsLayer::function, Layer::h, Net::head, Llist::layer, layer_args_init(), layer_init(), Layer::max_outputs, ArgsLayer::momentum, Layer::n_inputs, ArgsLayer::n_inputs, Net::n_layers, Layer::n_outputs, PredNeural::net, neural_insert(), neural_resize(), Llist::next, Cl::pred, and ArgsLayer::sgd_weights.

Referenced by xcsf_pred_expand().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred_neural_free()

void pred_neural_free	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Frees the memory used by a neural network prediction.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is to be freed.

Definition at line 58 of file pred_neural.c.

References PredNeural::net, neural_free(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_init()

void pred_neural_init	(	const struct XCSF *	xcsf,
		struct Cl *	c
	)

Creates and initialises a neural network prediction.

Uses fully-connected layers.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is to be initialised.

Definition at line 45 of file pred_neural.c.

References neural_create(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_json_export()

char* pred_neural_json_export	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Returns a json formatted string representation of a neural prediction.

Parameters

[in]	xcsf	XCSF data structure.
[in]	c	Classifier whose prediction is to be returned.

Returns: String encoded in json format.

Definition at line 392 of file pred_neural.c.

References PredNeural::net, neural_json_export(), and Cl::pred.

Referenced by pred_neural_print().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred_neural_json_import()

void pred_neural_json_import	(	const struct XCSF *	xcsf,
		struct Cl *	c,
		const cJSON *	json
	)

Creates a neural prediction from a cJSON object.

Parameters

[in]	xcsf	The XCSF data structure.
[in,out]	c	The classifier to initialise.
[in]	json	cJSON object.

Definition at line 414 of file pred_neural.c.

References PredNeural::net, neural_json_import(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_layers()

int pred_neural_layers	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Returns the number of layers within a neural network prediction.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	Classifier maintaining a neural network prediction.

Returns: The number of layers.

Definition at line 292 of file pred_neural.c.

References Net::n_layers, PredNeural::net, and Cl::pred.

Referenced by clset_mean_pred_layers().

Here is the caller graph for this function:

◆ pred_neural_load()

size_t pred_neural_load	(	const struct XCSF *	xcsf,
		struct Cl *	c,
		FILE *	fp
	)

Reads a neural network prediction from a file.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is to be read.
[in]	fp	Pointer to the file to be read.

Returns: The number of elements read.

Definition at line 199 of file pred_neural.c.

References neural_load(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_mutate()

bool pred_neural_mutate	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Mutates a neural network prediction.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is being mutated.

Returns: Whether any alterations were made.

Definition at line 153 of file pred_neural.c.

References PredNeural::net, neural_mutate(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_neurons()

int pred_neural_neurons	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const int	layer
	)

Returns the number of neurons in a neural prediction layer.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	Classifier maintaining a neural network prediction.
[in]	layer	Position of a layer in the network.

Returns: The current number of neurons in a layer.

Definition at line 240 of file pred_neural.c.

References CONVOLUTIONAL, Llist::layer, Layer::n_filters, Layer::n_outputs, PredNeural::net, Cl::pred, Llist::prev, Net::tail, and Layer::type.

Referenced by clset_mean_pred_neurons().

Here is the caller graph for this function:

◆ pred_neural_param_defaults()

void pred_neural_param_defaults ( struct XCSF * xcsf )

Initialises default neural prediction parameters.

Parameters

[in] xcsf The XCSF data structure.

Definition at line 431 of file pred_neural.c.

References CONNECTED, ArgsLayer::eta, ArgsLayer::evolve_connect, ArgsLayer::evolve_eta, ArgsLayer::evolve_neurons, ArgsLayer::evolve_weights, ArgsLayer::function, layer_args_copy(), layer_args_init(), LOGISTIC, ArgsLayer::max_neuron_grow, ArgsLayer::momentum, ArgsLayer::n_init, ArgsLayer::n_inputs, ArgsLayer::n_max, ArgsLayer::next, ArgsLayer::sgd_weights, and ArgsLayer::type.

Referenced by pred_param_defaults().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred_neural_param_json_import()

char* pred_neural_param_json_import	(	struct XCSF *	xcsf,
		cJSON *	json
	)

Sets the neural network parameters from a cJSON object.

Parameters

[in,out]	xcsf	The XCSF data structure.
[in]	json	cJSON object.

Returns: NULL if successful; or the name of parameter if not found.

Definition at line 465 of file pred_neural.c.

References layer_args_free(), layer_args_init(), layer_args_json_import(), layer_args_validate(), ArgsLayer::n_inputs, and ArgsLayer::next.

Referenced by pred_param_json_import().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred_neural_print()

void pred_neural_print	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Prints a neural network prediction.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is to be printed.

Definition at line 122 of file pred_neural.c.

References pred_neural_json_export().

Here is the call graph for this function:

◆ pred_neural_save()

size_t pred_neural_save	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		FILE *	fp
	)

Writes a neural network prediction to a file.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction is to be written.
[in]	fp	Pointer to the file to be written.

Returns: The number of elements written.

Definition at line 183 of file pred_neural.c.

References PredNeural::net, neural_save(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_size()

double pred_neural_size	(	const struct XCSF *	xcsf,
		const struct Cl *	c
	)

Returns the size of a neural network prediction.

See also: neural_size()

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	The classifier whose prediction size to return.

Returns: The network size.

Definition at line 168 of file pred_neural.c.

References PredNeural::net, neural_size(), and Cl::pred.

Here is the call graph for this function:

◆ pred_neural_update()

void pred_neural_update	(	const struct XCSF *	xcsf,
		const struct Cl *	c,
		const double *	x,
		const double *	y
	)

Backward propagates and updates a neural network prediction.

Precondition: The prediction has been forward propagated for the current state.

Parameters

[in]	xcsf	The XCSF data structure.
[in]	c	Classifier whose prediction is to be updated.
[in]	x	Input state.
[in]	y	Truth/payoff value.

Definition at line 91 of file pred_neural.c.

References PredNeural::net, neural_learn(), and Cl::pred.

Here is the call graph for this function:

Functions

Detailed Description

Function Documentation

◆ pred_neural_ae_to_classifier()

◆ pred_neural_compute()

◆ pred_neural_connections()

◆ pred_neural_copy()

◆ pred_neural_crossover()

◆ pred_neural_eta()

◆ pred_neural_expand()

◆ pred_neural_free()

◆ pred_neural_init()

◆ pred_neural_json_export()

◆ pred_neural_json_import()

◆ pred_neural_layers()

◆ pred_neural_load()

◆ pred_neural_mutate()

◆ pred_neural_neurons()

◆ pred_neural_param_defaults()

◆ pred_neural_param_json_import()

◆ pred_neural_print()

◆ pred_neural_save()

◆ pred_neural_size()

◆ pred_neural_update()