1lnn.h File Reference

Machine learning functionality for a 1-layer neural network. More...

#include <stdio.h>

Go to the source code of this file.

Data Structures
struct	Cell
	Core unit of the neural network (neuron and synapses) More...
struct	Layer
	The single (output) layer of this network (a layer is number cells) More...
struct	Vector
	Data structure containing defined number of integer values (the output vector contains values for 0-9) More...

Macros
#define	NUMBER_OF_INPUT_CELLS   784
	use 28*28 input cells (= number of pixels per MNIST image) More...
#define	NUMBER_OF_OUTPUT_CELLS   10
	use 10 output cells to model 10 digits (0-9) More...
#define	LEARNING_RATE   0.05
	Incremental increase for changing connection weights. More...

Typedefs
typedef struct Cell	Cell
typedef struct Layer	Layer
typedef struct Vector	Vector

Functions
Vector	getTargetOutput (int targetIndex)
	Returns an output vector with targetIndex set to 1, all others to 0. More...
void	initLayer (Layer *l)
	Initialize layer by setting all weights to random values [0-1]. More...
int	getLayerPrediction (Layer *l)
	Returns the index of the cell with the hightest output. More...
void	setCellInput (Cell *c, MNIST_Image *img)
	Sets a cell's input according to the pixels of a given MNIST image. More...
void	calcCellOutput (Cell *c)
	Calculates a cell's output by suming all input-weight-products. More...
double	getCellError (Cell *c, int target)
	Returns the difference between a target value and the cell's ouput. More...
void	updateCellWeights (Cell *c, double err)
	Updates a cell's weights based on given error and LEARNING_RATE. More...
void	trainCell (Cell *c, MNIST_Image *img, int target)
	Performs the training algorithm. More...
void	testCell (Cell *c, MNIST_Image *img, int target)
	Performs the testing of the trained network. More...

Detailed Description

Machine learning functionality for a 1-layer neural network.

Author

Matt Lind

Date

July 2015

Definition in file 1lnn.h.

Macro Definition Documentation

#define LEARNING_RATE   0.05

Incremental increase for changing connection weights.

Definition at line 14 of file 1lnn.h.

#define NUMBER_OF_INPUT_CELLS   784

use 28*28 input cells (= number of pixels per MNIST image)

Definition at line 11 of file 1lnn.h.

#define NUMBER_OF_OUTPUT_CELLS   10

use 10 output cells to model 10 digits (0-9)

Definition at line 12 of file 1lnn.h.

Typedef Documentation

typedef struct Cell Cell

Definition at line 18 of file 1lnn.h.

typedef struct Layer Layer

Definition at line 19 of file 1lnn.h.

typedef struct Vector Vector

Definition at line 20 of file 1lnn.h.

Function Documentation

void calcCellOutput

(

Cell *

c

)

Calculates a cell's output by suming all input-weight-products.

Parameters

c	A cell of a NN layer

Calculates a cell's output by suming all input-weight-products and normalizes to [0-1].

Definition at line 104 of file 1lnn.c.

double getCellError	(	Cell *	c,
		int	target
	)

Returns the difference between a target value and the cell's ouput.

Parameters

c	The cell whose output is to be compared
target	The desired value (= correct answer in supervised learning)

Returns the difference between a target value and the cell's ouput

Definition at line 122 of file 1lnn.c.

int getLayerPrediction

(

Layer *

l

)

Returns the index of the cell with the hightest output.

Parameters

l	A pointer to a NN layer

The output prediction is derived by simply sorting all output values and using the index (=0-9 number) of the highest value as the prediction.

Definition at line 62 of file 1lnn.c.

Vector getTargetOutput

(

int

targetIndex

)

Returns an output vector with targetIndex set to 1, all others to 0.

Parameters

targetIndex

Index of the output that is to be set to 1

Returns an output vector with targetIndex set to 1, all others to 0

Definition at line 22 of file 1lnn.c.

void initLayer

(

Layer *

l

)

Initialize layer by setting all weights to random values [0-1].

Parameters

l	A pointer to a NN layer

Initialize layer by setting all weights to random values [0-1]

Attention

It actually makes no difference whether the weights are initialized to a constant (e.g. 0.5) or to a random number. The result (85% success rate) will not change significantly.

Definition at line 40 of file 1lnn.c.

void setCellInput	(	Cell *	c,
		MNIST_Image *	img
	)

Sets a cell's input according to the pixels of a given MNIST image.

Parameters

c	A pointer to a cell
img	A pointer to an MNIST image

Creates an input vector of length NUMBER_OF_INPUT_CELLS of a given MNIST image, setting input vector cells to [0,1] based on the pixels of the image. Scalar pixel intensity [=grey-scale] is ignored, only 0 or 1 [=black-white].

Definition at line 89 of file 1lnn.c.

void testCell	(	Cell *	c,
		MNIST_Image *	img,
		int	target
	)

Performs the testing of the trained network.

Parameters

c	Pointer to the cell that is to be trained
img	Pointer to the image that is to be processed
target	Desired output value

Performs the testing of the trained network Same as training a cell, but without updating weights (learning)

Definition at line 169 of file 1lnn.c.

void trainCell	(	Cell *	c,
		MNIST_Image *	img,
		int	target
	)

Performs the training algorithm.

Parameters

c	Pointer to the cell that is to be trained
img	Pointer to the image that is to be processed
target	Desired output value

Performs the training algorithm: feeding input, calculate output, calculate error, update weights)

Definition at line 151 of file 1lnn.c.

void updateCellWeights	(	Cell *	c,
		double	err
	)

Updates a cell's weights based on given error and LEARNING_RATE.

Parameters

c	The cell whose weights are to be updated.
err	The error (difference between desired output and actual output

Updates a cell's weights based on given error and LEARNING_RATE

Definition at line 136 of file 1lnn.c.