BP algorithm-written in Java

// BP neural network algorithm implementation </P> <p> # include <stdio. h> <br/> # include <math. h> <br/> # include <conio. h> <br/> # include <stdlib. h> </P> <p> # define NH 3/* input layer */<br/> # define Ni 4/* hidden layer */<br/> # define NJ 1 /* output layer */<br/> # define NK 100/* Sample size */<br/> # define test 70/* Test Set capacity */<br/> # define NR 0.7/* learning efficiency */<br/> # define EPS 0.00001 </P> <p> float X [NK] [NH], d [NK] [NJ], whi [NH] [Ni], wij [Ni] [NJ], Thi [Ni], THJ [NJ]; <br/> int H, I, J, K, ff; <br/> fl Oat xmin [NH], xmax [NH], dmin [NJ], dmax [NJ]; <br/> file * FP1, * fp2, * fp3, * fp4; </P> <p> void Init (void); <br/> void startleaning (void); <br/> void testsample (void ); <br/> void readw (void); <br/> void readt (void); <br/> void writew (void ); <br/> float sigmoid (float a); <br/> double ranu (void); <br/> char filename1 [] = {"samplefile.txt "}; </P> <p> void Init (void) <br/> {<br/> int min, Max; <br/> If (FP1 = 0) <br/>{< br/> System ("CLS"); <br/> printf ("can not find the learning sample file! /N "); <br/> exit (0); <br/>}< br/> for (k = 0; k <NK; k ++) <br/> {<br/> for (H = 0; H <nH; H ++) <br/> fscanf (FP1, "% F ,", & X [k] [H]); // neural network input <br/> for (j = 0; j <NJ; j ++) <br/> fscanf (FP1, "% F,", & D [k] [J]); // Neural Network output <br/>}< br/> for (H = 0; H <nH; H ++) <br/> {<br/> min = 1; <br/> max = 1; <br/> for (k = 0; k <NK; k ++) <br/> {<br/> If (X [k] [H] <X [Min] [H]) min = K; <br/> If (X [k] [H]> X [Max] [H]) max = K; <br/>}< br/> xmin [H] = x [Min] [H]; <br/> xmax [H] = x [Max] [H]; <Br/> for (k = 0; k <NK; k ++) /* neural network input normalization */<br/> X [k] [H] = (X [k] [H]-xmin [H]) /(xmax [H]-xmin [H]); <br/>/* X [k] [H] = x [k] [H]/xmax [H]; */<br/>}< br/> for (j = 0; j <NJ; j ++) <br/>{< br/> min = 1; max = 1; <br/> for (k = 0; k <NK; k ++) <br/> {<br/> If (d [k] [J] <D [Min] [J]) <br/> min = K; <br/> If (d [k] [J]> d [Max] [J]) <br/> max = K; <br/>}< br/> dmin [J] = d [Min] [J]; <br/> dmax [J] = d [Max] [J]; <br/> for (k = 0; k <NK; k ++) /* Neural Network output normalization */<br/> d [k] [J] = (d [k] [J]-DMI N [J])/(Dmax [J]-dmin [J]); <br/>/* d [k] [J] = d [k] [J]/dmax [J]; */<br/>}< br/>/* ---------------------------------------------------- */<br/> void startlearning (void) <br/>{< br/> long int nt, N; <br/> float T, error [NK], gerror, XJ [NJ], Xi [Ni], YJ [NJ], Yi [Ni], PXI [Ni], pxj [NJ]; <br/> float U0 = 0, u1 = 0, U2 = 0, u3 = 0; <br/> float v0, V1, V2, V3; </P> <p> for (I = 0; I <Ni; I ++) <br/> {<br/> for (H = 0; H <nH; H ++) <br/> whi [H] [I] =-0.8 + 1.6 * Ranu (); <br/> for (j = 0; j <NJ; j ++) <br/> wij [I] [J] =-0.8 + 1.6 * ranu (); </P> <p> Thi [I] =-0.5 + ranu (); <br/>}< br/> for (j = 0; j <NJ; j ++) <br/> THJ [J] =-0.5 + ranu (); <br/> fp2 = fopen ("cmdtxt", "W + "); /* initialization of weights and thresholds */</P> <p>/* Start of learning */<br/> printf ("/T/nplease enter the learning times: /n "); <br/> scanf (" % lD ", & NT); <br/> for (n = 0; n <NT; n ++) /* learning times */<br/> {<br/> gerror = 0; <br/> for (k = 0; k <NK; k ++) /* single sample loop */<br/> {<br/> for (I = 0; I <Ni; I ++) <br/> {<Br/> T = 0; <br/> for (H = 0; H <nH; H ++) <br/> T + = whi [H] [I] * X [k] [H]; <br/> Xi [I] = T + Thi [I]; <br/> Yi [I] = sigmoid (XI [I]);/* hidden layer output */<br/>}< br/> for (j = 0; j <NJ; j ++) <br/>{< br/> T = 0; <br/> for (I = 0; I <Ni; I ++) <br/> T + = wij [I] [J] * Yi [I]; <br/> XJ [J] = T + THJ [J]; <br/> YJ [J] = sigmoid (XJ [J]); <br/>}/ * output layer */</P> <p> for (j = 0; j <NJ; j ++) /* error change rate of single sample points in the output layer */<br/> pxj [J] = YJ [J] * (1-yj [J]) * (YJ [J]-d [k] [J]); <br/> for (I = 0; I <Ni; I ++) /* hidden layer single sample point error change rate * /<Br/> {<br/> T = 0; <br/> for (j = 0; j <NJ; j ++) <br/> T + = pxj [J] * wij [I] [J]; <br/> PXI [I] = Yi [I] * (1-yi [I]) * t; <br/>}< br/> for (j = 0; j <NJ; j ++) <br/> {<br/> // THJ [J] = THJ [J]-Nr * pxj [J]; <br/> V0 = THJ [J]; <br/> THJ [J] = THJ [J]-0.7 * Nr * pxj [J] + 0.3 * U0; <br/> // THJ [J] = THJ [J]-Nr * pxj [J]; <br/> U0 = THJ [J]-V0; <br/> for (I = 0; I <Ni; I ++) <br/> {<br/> V1 = wij [I] [J]; <br/> wij [I] [J] = wij [I] [J]-0.7 * Nr * pxj [J] * Yi [I] + 0.3 * U1; <br/> U1 = wij [I] [J]-V1; <Br/> // wij [I] [J] = wij [I] [J]-Nr * pxj [J] * Yi [I]; /* modify the weight from the hidden layer to the output layer, where NR is the step size */<br/>}< br/> for (I = 0; I <Ni; I ++) <br/>{< br/> v2 = Thi [I]; <br/> Thi [I] = Thi [I]-0.7 * Nr * PXI [I] + 0.3 * U2; <br/> u2 = Thi [I]-V2; <br/> // Thi [I] = Thi [I]-Nr * PXI [I]; <br/> for (H = 0; H <nH; h ++) <br/>{< br/> V3 = whi [H] [I]; <br/> whi [H] [I] = whi [H] [I]-0.7 * Nr * PXI [I] * X [k] [H] + 0.3 * U3; <br/> U3 = whi [H] [I]-V3; <br/> // whi [H] [I] = whi [H] [I]-Nr * PXI [I] * X [k] [H]; /* input layer Hidden Layer weight correction, where NR is the step */<br/>}< br/> T = 0; <br/> for (j = 0; j <NJ; j ++) <br/> T + = (YJ [J]-d [k] [J]) * (YJ [J]-d [k] [J])/2.0; <br/> error [k] = T; <br/> gerror + = Error [k];/* Global error g (lobal) error */<br/>}< br/>/* single sample cycle ends */</P> <p> If (gerror <EPS) <br/> break; <br/>}< br/>/* learning cycle ends */<br/> printf ("% F, % F ", Thi [0], THJ [0], wij [0] [0], wij [0] [1], whi [1] [0], whi [0] [1]); <br/> writew (); <br/> printf ("/T/nglobal error = % F/N", gerror ); <Br/> printf ("/T/Nare you satisfied with the global error? /N "); <br/> printf (" press any key to choose a next task! /N "); <br/> getch (); <br/>}< br/>/* ----------------------------------------------- */<br/> void testsample (void) <br/> {<br/> float TX [NH], T, XJ [NJ], Xi [Ni], YJ [NJ], Yi [Ni]; <br/> If (fp2 = 0) <br/>{< br/> // clrscr (); <br/> printf ("/T/ncan not find the weight file: cmdtxt/N"); <br/> exit (0 ); <br/>}< br/> readw (); </P> <p> for (FF = 0; FF <test; FF ++) /* test sample prediction and result storage */<br/> {<br/> for (H = 0; H <nH; H ++) <br/> fscanf (fp3, "% F ,",& TX [H]); <br/> for (H = 0; H <nH; H ++) <br/> TX [H] = (TX [H]-xmin [H])/(xmax [H]-xmin [H]); /* normalization process */<br/> for (I = 0; I <Ni; I ++) <br/> {<br/> T = 0; <br/> for (H = 0; H <nH; H ++) <br/> T + = whi [H] [I] * TX [H]; <br/> Xi [I] = T + Thi [I]; <br/> Yi [I] = sigmoid (XI [I]); <br/>}< br/> for (j = 0; j <NJ; j ++) <br/> {<br/> T = 0; <br/> for (I = 0; I <Ni; I ++) <br/> T + = wij [I] [J] * Yi [I]; <br/> XJ [J] = T + THJ [J]; <br/> YJ [J] = sigmoid (XJ [J]); <br/>}< br/> // printf ("/T/NNetwork Output:/N ");/* Forward prediction */<br/> for (j = 0; j <NJ; j ++) /* Save the result */<br/> {<br/> YJ [J] = YJ [J] * (Dmax [J]-dmin [J]) + dmin [J]; <br/> fprintf (fp4, "% F/N", YJ [J]); <br/>}< br/> // TX [nh-1] = YJ [0]; <br/> // rewind (fp4 ); <br/> // For (H = 0; H <nH; H ++)/* test sample input */<br/> // fscanf (fp3, "% F,", & TX [H]); <br/>}</P> <p> printf ("/T/Nare you satisfied with the output? /N "); <br/> printf (" press any key to choose a next task! /N "); <br/> getch (); <br/>}< br/>/* ---------------------------------------------- */<br/> void writew (void) <br/>{< br/> rewind (fp2); <br/> for (H = 0; H <nH; H ++) <br/> {<br/> for (I = 0; I <Ni; I ++) <br/> fprintf (fp2, "% 8.3f ", whi [H] [I]); <br/> fprintf (fp2, "/N"); <br/>}< br/> fprintf (fp2, "/N"); </P> <p> for (I = 0; I <Ni; I ++) <br/> fprintf (fp2, "% 8.3f ", thi [I]); <br/> fprintf (fp2, "/n"); </P> <p> for (j = 0; j <NJ; j ++) <br/>{< br/> for (I = 0; I <Ni; I ++) <br/> fprintf (fp2, "% 8.3f ", wij [I] [J]); <br/> fprintf (fp2, "/N"); <br/>}< br/> fprintf (fp2, "/N"); <br/> for (j = 0; j <NJ; j ++) <br/> fprintf (fp2, "% 8.3f ", THJ [J]); <br/>}< br/>/* ------------------------------------------------ */<br/> void readw (void) <br/> {<br/> for (H = 0; H <nH; H ++) <br/> for (I = 0; I <Ni; I ++) <br/> fscanf (fp2, "% F", & whi [H] [I]); <br/> for (I = 0; I <Ni; I ++) <br/> fscanf (fp2, "% F", & Thi [I]); <br/> for (j = 0; j <NJ; j ++) <br/> for (I = 0; I <Ni; I ++) <br/> fscanf (fp2, "% F ", & wij [I] [J]); <br/> for (j = 0; j <NJ; j ++) <br/> fscanf (fp2, "% F", & THJ [J]); <br/>}< br/>/* -------------------------------- */<br/> float sigmoid (float) <br/>{< br/> return (1.0/(1 + exp (-))); <br/>}< br/>/* ------------------------------ */<br/> double ranu (void) <br/>{< br/> static double xrand = 3.0; <br/> double M = 8589934592.0, A = 30517578125.0; <br/> LP: xrand = fmod (xrand * a, m ); /* remainder Division */<br/> If (xrand> 1.0) <br/> return (xrand/M ); <br/> else <br/> {<br/> xrand = 1.0; <br/> goto LP; <br/>}< br/>/* ---------------------------------- */</P> <p> void main () <br/>{< br/> FP1 = fopen ("sample.txt", "R"); <br/> fp2 = fopen ("cmdtxt", "R + "); <br/> fp3 = fopen ("test.txt", "R +"); <br/> fp4 = fopen ("testre.txt", "R + "); <br/> Init (); <br/> while (1) <br/> {<br/> system ("CLS "); <br/> printf ("/T/N please choose a next task... /n "); <br/> printf ("/T/N (s) to start learning. /n "); <br/> printf ("/T/N (t) to test samples. /n "); <br/> printf ("/T/N (r) to resume learning. /n "); <br/> printf ("/T/N (q) quit. /n "); <br/> switch (getchar () <br/>{< br/> case 's': startlearning (); break; <br/> case 'T': testsample (); break; <br/> case 'r': startlearning (); break; <br/> case 'q ': exit (0); break; <br/>}< br/> fclose (FP1); <br/> fclose (fp2 ); <br/> fclose (fp3); <br/> fclose (fp4); <br/> getch (); <br/>}< br/>/* ---------------- by ahzhming@163.com sea Wide Sky ------------------------- */<br/>