55 -14.3693 -13.4362 -11.4072 -12.3129 -14.8612 -13.3480 -12.8517

-13.4014 -0.7738 0
56 -14.2856 -12.6858 -13.8215 -13.4282 -14.0982 -13.1587 -13.2792

-13.7852 -1.3442 0
57 -14.4822 -13.1141 -13.7787 -13.4466 -13.6761 -13.2969 -13.6033

-13.9252 -0.6642 1
58 -13.5522 -13.1302 -13.5444 -14.1471 -13.2994 -13.2368 -13.9776

-14.4295 -0.9973 1
59 -14.8524 -11.9846 -13.7231 -14.2496 -13.4809 -13.0515 -13.8950

-14.3923 -1.8284 1
…

85 -14.5994 -13.6920 -12.8539 -13.7629 -14.1699 -13.2075 -13.3422

-13.6788 -11.9537 1
86 -14.3821 -13.6093 -12.8677 -13.7788 -14.1260 -13.3246 -13.2966

-13.6453 -11.4304 1

3. Файл с настройками программы

#
# Common parameters for programm "NVCLASS"
#
# # # # # # # # # # # # # # # # # # # #
# 1_1 - OnlyTest mode , 1_2 - TestAfterLearn mode,
# 2_1 - CheckOneVector , 2_2 - CrossValidation mode.
#
TYPE=2_2
NDATA=9
NPATTERN=86
PatternFile=9_Norv.txt
NTEST=10
TestVector=vector.tst
NetworkFile=9.net
ResNetFname=9.net
NumberVector=57
ReportFile=Report.txt
Debug=Yes
#
# Next parameters was define in result experiments and if you will
# change it, the any characteristics of Neural Net may be not optimal
# (since may be better then optimal).

#
# # # # # # # # # # #
# 'NetStructure' must be: [NDATA,NUNIT1,1] (NOUT=1 always)
# value 'AUTO'-'NetStructure' will be define the programm.(See help).
# example : [18,9,1], or [18,18,1], or [9,9,5,1]
NetStructure=[18,12,1]
# may be: [Gauss] or [Random]
InitWeigthFunc=Gauss
Constant=3
Alfa=0
Sigma=1.5
Widrow=No
Shuffle=Yes
Scaling=Yes
LearnTolerance=0.1

Eta=1

MaxLearnCycles=50
Loop=3
#end of list

4. Пример файла отчета.

NVCLASS report - Wed Jun 02 15:58:02 1999

Type = 1_2

Neural Net -

PatternFile - vect.txt

Test Vector(s) - vector.tst

ResNetFname - 12.net

LearnTolerance = 0.10

InitialWeigthFunc = Gauss[ 0.0, 1.5]

< Loop 1 >
Learning cycle result:

NumIter = 5

NumLE = 3

Error vector(s): 58, 59, 63,
+-----+------+--------+------+
| N | ID | Result |Target|
+-----+------+--------+------+
| 1 | 24 | 0.1064 | 0 |
| 2 | 25 | 0.9158 | 1 |
| 3 | 26 | 0.0452 | 0 |
| 4 | 27 | 0.0602 | 0 |
| 5 | 28 | 0.0348 | 0 |
| 6 | 29 | 0.0844 | 0 |
| 7 | 30 | 0.1091 | 0 |
| 8 | 31 | 0.0821 | 0 |
| 9 | 32 | 0.0298 | 0 |
| 10 | 33 | 0.2210 | 0 |
+-----+------+--------+------+

< Loop 2 >
Learning cycle result:

NumIter = 5

NumLE = 5

Error vector(s): 33, 34, 55, 58, 63,
+-----+------+--------+------+
| N | ID | Result |Target|
+-----+------+--------+------+
| 1 | 24 | 0.1279 | 0 |
| 2 | 25 | 0.9929 | 1 |
| 3 | 26 | 0.0960 | 0 |
| 4 | 27 | 0.1463 | 0 |
| 5 | 28 | 0.1238 | 0 |
| 6 | 29 | 0.1320 | 0 |
| 7 | 30 | 0.1478 | 0 |
| 8 | 31 | 0.1235 | 0 |
| 9 | 32 | 0.0740 | 0 |
| 10 | 33 | 0.5140 | 1 |
+-----+------+--------+------+

5. Файл описания функций, типов переменных и используемых библиотек

“nvclass.h”.
/*
* --- Neuro classificator---
*
* Common defines
*/

#include
#include
#include
#include
#include
#include
#include
//#include
//#include
#include

#define DefName "nvclass.inp"

#define MAXDEF 100
#define MAXLINE 256
#define NMAXPAT 100
#define NMXINP 20
#define NMXUNIT 20

#define CONT 0

#define EXIT_OK 1
#define EXIT_CNT 2

#define RESTART 911
#define MAXEXP 700 /* Max arg exp(arg) without error 'OVERFLOW'
*/

#define Random 10
#define Gauss 20

#define OK 0
#define Error 1
#define Yes 77
#define No 78
#define Min 0 /* Find_MinMax(...) */
#define Max 1

#define TYPE_ONE 21
#define TYPE_TWO 22
#define TYPE_THREE 23
#define TYPE_FOUR 24

int NDATA = 0; int NUNIT1 = 0; int NUNIT2 = 0; int NUNIT3 = 0; int NOUT = 1;

int NPATTERN = 0; /* Number of input pattern*/ int NWORK = 0; /* Number of work pattern*/ int NTEST= 0; /* Number of test pattern*/

int result; int STOP = 0;

int NumOut = 250; /* Number of itteration, after which show result in debugfile. */ int Num_Iter=10;/* The parameters requred in the procecc of */ float Percent=0.25; /* dinamic lerning with change 'eta' */

float LearnTolerance = 0.10; float TestTolerance = 0.5;

float MAX_ERR=0.00001; /* min error */ float eta = 1.0; /* learning coefficient*/ float MIN_ETA=0.000001;

float **Array_MinMax; int *Cur_Number;

float W1[NMXINP][NMXUNIT]; float W2[NMXUNIT];

float PromW1[NMXINP][NMXUNIT]; float PromW2[NMXUNIT];

float PromW1_OLD[NMXINP][NMXUNIT]; float PromW2_OLD[NMXUNIT];

float Err1[NMXUNIT]; float Err2; float OLD_ERROR; float GL_Error=0.0;

float Out1[NMXUNIT]; float Out2;

char NetStr[20]="Auto"; /* String with pattern of Net Structure*/

int Type = TYPE_THREE; /* Enter the mode of work of programm */

int InitFunc = Random; /* Random [=10] weigth will RandomDistribution
Gauss [=20] - ... GaussianDistributon */ float Constant = 1; /* RandomDistribution [-Constant,Constant]*/

float Alfa = 0; /* GaussianDistribution [Alfa,Sigma]*/ float Sigma = 1; /* ... */ int Widrow = No; /* Nguyen-Widrow initialization start weigth*/

int Loop = 1; /* Number repeat of Learning cycle */

char *PatternFile; /* File with input patterns*/ char *TestVector; char *ReportFile="report.txt"; /* name of report file */ char *NetworkFile; /* Name of input NetConfig file */ char *ResNetFname; /* Name of output NetConfig file */

int DEBUG = Yes; /* if 'Yes' then debug info in the DebugFile
*/ char *DebugFile="Logfile.log"; /* Name of the debug file*/

int NumberVector = 0; /* Number of TEST vector */ int Shuffle = Yes; /* Flag - shuffle the input vectors*/ int Scaling = Yes; /* Scaling input vector */ int MaxLearnCycles = 1999; /* Max number of learning iteration */

FILE *Dfp; /* Debug file pointer */
FILE *Rfp; /* Report file pointer*/

typedef struct Pattern { int ID; /* ID number this vector in all set of pattern */ float *A; /* pattern (vector) A={a[0],a[1],...,a[NDATA]} */ float Target; /* class which this vector is present*/
} PAT;

PAT *Input;
PAT *Work;
PAT *Test;

/* lines in defaults file are in the form "NAME=value" */ typedef struct Default { char *name; /* name of the default */ char *value; /* value of the default */
} DEF;

/* structure of statistics info about one test vector */ typedef struct Statistic { int ID; /* Primery number from input file */ float Target; float TotalRes; /* Total propability */ int Flag; /* Flag = 1, if vector was error and = 0 in over case */ float *result; /* Result of testing vector on current iteration */ int *TmpFlag; /* analog 'Flag' on current itteration */ int *NumIter; /* Number iteration of learning on which

Learning cycle STOPED */ int **NumLE; /* Error vectors after cycle of learning was test*/
} STAT;

/* structure of the some result of learning cycle */ typedef struct ResLearning { int NumIter; int LearnError[NMAXPAT+1]; /* A[0]-count of error,

A[1]-ID1,

A[2]-ID2,...

A[NMAXRL]-ID?.*/
} RL;

/* function prototypes */

void OnlyTestVector(void); void TestAfterLearn (void); void CheckOneVector ( void ); void CrossValidation ( void );

DEF **defbuild(char *filename);
DEF *defread(FILE *fp);
FILE *defopen (char *filename); char *defvalue(DEF **deflist, const char *name); int defclose(FILE *fp); void defdestroy(DEF **, int); void getvalues(void);

void Debug (char *fmt, ...); void Report (char *fmt, ...);

void Widrow_Init(void); int Init_W( void ); float RavnRaspr(float A, float B); float NormRaspr(float B,float A);

void ShufflePat(int *INP,int Koll_El);

float F_Act(float x); float Forward (PAT src); int LearnFunc (void); int Reset (float ResErr, int Cnt, int N_Err); void Update_Last (int n, float Total_Out); void Update_Prom1 (int n); void Prom_to_W (void); void Update_All_W (int num, float err_cur ); void Init_PromW(void); void Prom_to_OLD(void); int CheckVector(float Res, PAT src); int *TestLearn(int *src);

RL FurtherLearning(int NumIteration, float StartLearnTolerans, float EndLearnTolerans,

RL src);

STAT *definestat (PAT src);
STAT **DefineAllStat (PAT *src,int Num); void FillStatForm (STAT *st, int iteration, float res, RL lr); void FillSimpleStatForm (STAT *st, float res); void destroystat ( STAT *st, int param); void DestroyAllStat (STAT **st, int Num); void PrintStatHeader(void); void printstat(STAT *st); void PrintStatLearn(RL src); void PrintTestStat(STAT **st, int len); void PrintErrorStat (STAT **st,int Len);

int DefineNetStructure (char *ptr); void getStructure(char buf[20]);

PAT patcpy (PAT dest, PAT src);
PAT* LocPatMemory(int num); void ReadPattern (PAT *input, char *name,int Len); void FreePatMemory(PAT* src, int num); void ShowPattern (char *fname, PAT *src, int len); void ShowVector(char *fname,PAT src); float getPatTarget (float res);

PAT* DataOrder (PAT* src,int Len, int Ubit, PAT* dest, PAT* test); void FindMinMax (PAT *src,int Dimens, int Num_elem, float **Out_Array); void ConvX_AB_01(PAT src);

int *DefineCN (int len); int getPosition (int Num, int *src, int Len); void DestroyCN (int *src); void ShowCurN (int LEN);

float **LocateMemAMM(void); void FreeAMM (float **src);

void WriteHeaderNet(char *fname, float **src); void WriteNet (char *fname,int It); void ReadHeaderNet(char *fname, float **src); int ReadNet (char *fname, int It);
FILE *OpenFile(char *name); int CloseFile(FILE *fp);

/* End of common file */

6. Файл автоматической компиляции программы под Unix -“Makefile”.
CC= cc
LIBS= -lm

OBJ= nvclass.o

nvclass: $(OBJ)

$(CC) -o nvclass $(LIBS) $(OBJ)

nvclass.o: nvclass.c

7. Основной модуль - “nvclass.с”
/*
* Neuron Classificator ver 1.0
*/

#include "common.h"

/* =========================
* MAIN MODULE
* =========================
*/ void main (int argc, char *argv[])
{ int i; char buf[MAXLINE], PrName[20], *ptr; time_t tim; time(&tim);

/* UNIX Module */

Dfp = OpenFile(DebugFile); strcpy(buf,argv[0]); ptr = strrchr(buf,'/'); ptr++; strcpy(PrName,ptr);
Debug ("nn'%s' - Started %s",PrName,ctime(&tim));

getvalues();
Rfp = OpenFile(ReportFile);

DefineNetStructure(NetStr); /* NetStr string from input file */ getStructure(buf);

Debug ("nNeyral net %s",buf);
Input = LocPatMemory(NPATTERN);
Work = LocPatMemory(NPATTERN);

Array_MinMax = LocateMemAMM();
Cur_Number = DefineCN (NPATTERN);

printf("nMetka - 1"); if (Type == TYPE_ONE)

OnlyTestVector (); if (Type == TYPE_TWO)

TestAfterLearn (); if (Type == TYPE_THREE)

CheckOneVector (); if (Type == TYPE_FOUR)

CrossValidation();

time(&tim);
Debug ("nn%s - Normal Stoped %s",PrName,ctime(&tim));

CloseFile(Dfp);
CloseFile(Rfp);
FreeAMM (Array_MinMax);
DestroyCN (Cur_Number);
FreePatMemory(Input,NPATTERN);
FreePatMemory(Work, NPATTERN);
}

/*
* ^OnlyTestVectors - read net from (NetworkFile) and test the
TestVector(s)
*/ void OnlyTestVector(void)
{ char buf[MAXLINE+1];

STAT **st, *stat; int i,j; float Res;

Debug ("nOnlyTestVector proc start");
Debug ("n NPATTERN = %d",NPATTERN);
Debug ("n NTEST = %d",NTEST);

Test = LocPatMemory(NTEST);
ReadPattern(Test,TestVector, NTEST);
/* ShowPattern ("1.tst",Test,NTEST);*/
PrintStatHeader(); st = DefineAllStat (Test,NTEST);
ReadHeaderNet(NetworkFile,Array_MinMax); if (Scaling == Yes)

{ for (i=0;i ",i+1);

ReadNet(NetworkFile,i+1); for (j=0;j

Страницы: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10