http://iofai.com/204.html
BP神经网络演示类-Delphi版
这是一个以前写的一个Bp人工神经网络训练的类,感觉收敛的速度不错,发布出来供学习人工神经网络的朋友们参考:
001
002
003
004
005
006
007
008
009
010
011
012
013
014
015
016
017
018
019
020
021
022
023
024
025
026
027
028
029
030
031
032
033
034
035
036
037
038
039
040
041
042
043
044
045
046
047
048
049
050
051
052
053
054
055
056
057
058
059
060
061
062
063
064
065
066
067
068
069
070
071
072
073
074
075
076
077
078
079
080
081
082
083
084
085
086
087
088
089
090
091
092
093
094
095
096
097
098
099
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
|
unit UBP; {代码-智能办公http://iofai.com} interface uses Math,SysUtils; type ObjNeurons = object _U: Single ; _B: Single ; _PB: Single ; _Type: Byte ; _Offset: Single ; //at _SumOffset: Single ; //sum at _OldSumOffset: Single ; //old function _V(): Single ; end ; RNeuronsOfLevel = record _NeruronsCount: Integer ; _NeruronsArray: array of ObjNeurons; //OBJ ARRAY end ; RSynaptic = record //突触 _Weight: Single ; _PW: Single ; _Offset: Single ; _SumOffset: Single ; _OldSumOffset: Single ; end ; RMatrix = record Matrix: array of array of RSynaptic; end ; RSample = record X: array of Single ; Y: array of Single ; end ; ObjNeuronsNet = class constructor create(); overload; constructor Create( const LevelArray: array of Integer ); overload; destructor Destroy(); override; public _SampleCount: Integer ; _SampleArray: array of RSample; _MaxStudyCount: Integer ; //最大学习次数 _MaxError: Single ; _UpN: Single ; //N上升 _DownN: Single ; //N下降 _MaxN: Single ; //最大N procedure MemSample(); //分配 procedure Study(); //day day Study haha private _StudyCount: Integer ; _LevelCount: Integer ; //Level Count I to Calc _WeightMatrix: array of RMatrix; //Synaptic Matrix _NeruronsLevelArray: array of RNeuronsOfLevel; // Nerurons Of level _ExportArray: array of Single ; procedure FreeNet(); //Free procedure Changle(); //权值 //----- //----- procedure InitNet( const LevelArray: array of Integer ); //init BP Net,get Array of Level procedure InitWeightMatrix(); //init Matrix of Weight is random procedure CalcExport(); //Calc to procedure CalcExportReturn(); //return procedure CalcOffset( const LevelID: Integer ; const NeuronsID: Integer ); //Neurons Offset procedure SaveOffset(); //save procedure AddOffset(); //Sum Offset function GetError(): Single ; end ; implementation uses UMain; { ObjNeurons } function ObjNeurons . _V: Single ; begin case _Type of 0 : begin Result := _U; end ; 1 : begin Result := 1 / ( 1 + Exp(-(_U - _B))); end ; end ; end ; { ObjNeuronsNet } procedure ObjNeuronsNet . AddOffset; var i, j, k: Integer ; Count: Integer ; ACount: Integer ; begin for i := 0 to _LevelCount - 1 do begin count := _NeruronsLevelArray[i]._NeruronsCount; Acount := _NeruronsLevelArray[i + 1 ]._NeruronsCount; for j := 0 to Count - 1 do begin _NeruronsLevelArray[i]._NeruronsArray[j]._SumOffset := _NeruronsLevelArray[i]._NeruronsArray[j]._SumOffset + _NeruronsLevelArray[i]._NeruronsArray[j]._Offset; if i < (_LevelCount - 1 ) then begin for k := 0 to ACount - 1 do begin _WeightMatrix[i].Matrix[J, K]._SumOffset := _WeightMatrix[i].Matrix[J, K]._SumOffset + _WeightMatrix[i].Matrix[J, K]._Offset; end ; end ; end ; end ; end ; procedure ObjNeuronsNet . CalcExport; var i, j, k: Integer ; FCount: Integer ; Sum: Single ; begin for i := 1 to _LevelCount - 1 do begin FCount := _NeruronsLevelArray[i - 1 ]._NeruronsCount; for j := 0 to _NeruronsLevelArray[i]._NeruronsCount - 1 do begin Sum := 0 ; for K := 0 to FCount - 1 do begin Sum := Sum + _WeightMatrix[i - 1 ].Matrix[K, J]._Weight * _NeruronsLevelArray[i - 1 ]._NeruronsArray[k]._V; end ; _NeruronsLevelArray[i]._NeruronsArray[j]._U := Sum; end ; end ; end ; procedure ObjNeuronsNet . CalcExportReturn; var i, j, k: Integer ; Count: Integer ; Acount: Integer ; begin for i := _LevelCount - 1 downto 1 do begin count := _NeruronsLevelArray[i]._NeruronsCount; for j := 0 to Count - 1 do begin CalcOffset(i, j); end ; end ; for i := 0 to _LevelCount - 2 do begin count := _NeruronsLevelArray[i]._NeruronsCount; Acount := _NeruronsLevelArray[i + 1 ]._NeruronsCount; for j := 0 to Count - 1 do begin for k := 0 to Acount - 1 do begin _WeightMatrix[i].Matrix[j, k]._Offset := Self . _NeruronsLevelArray[i]._NeruronsArray[j]._V * Self . _NeruronsLevelArray[i + 1 ]._NeruronsArray[K]._Offset; end ; end ; end ; end ; procedure ObjNeuronsNet . CalcOffset( const LevelID, NeuronsID: Integer ); var i: Integer ; YExport, DExport: Single ; Sum: Single ; ACount: Integer ; begin Sum := 0 ; {本段已经加入一个负号} YExport := _NeruronsLevelArray[LevelID]._NeruronsArray[NeuronsID]._V; if LevelID = (_LevelCount - 1 ) then begin DExport := _ExportArray[NeuronsID]; _NeruronsLevelArray[LevelID]._NeruronsArray[NeuronsID]._Offset := -YExport * ( 1 - YExport) * (DExport - YExport); end else begin ACount := _NeruronsLevelArray[LevelID + 1 ]._NeruronsCount; for i := 0 to ACount - 1 do begin Sum := Sum + _NeruronsLevelArray[LevelID + 1 ]._NeruronsArray[i]._Offset * _WeightMatrix[LevelID].Matrix[NeuronsID, i]._Weight; end ; _NeruronsLevelArray[LevelID]._NeruronsArray[NeuronsID]._Offset := -YExport * ( 1 - YExport) * Sum; end ; end ; constructor ObjNeuronsNet . create; begin end ; procedure ObjNeuronsNet . Changle; var i, j, k: Integer ; FCount: Integer ; ACount: Integer ; PB: Single ; oldPb: Single ; PW: Single ; OldPw: Single ; T: Single ; B: Single ; W: Single ; begin for i := 0 to _LevelCount - 1 do begin FCount := _NeruronsLevelArray[i]._NeruronsCount; for j := 0 to FCount - 1 do begin PB := _NeruronsLevelArray[i]._NeruronsArray[j]._SumOffset; oldPb := _NeruronsLevelArray[i]._NeruronsArray[j]._OldSumOffset; T := PB * OLDPB; if T > 0 then begin _NeruronsLevelArray[i]._NeruronsArray[j]._PB := _UpN * _NeruronsLevelArray[i]._NeruronsArray[j]._PB; end else if (T < 0 ) then begin _NeruronsLevelArray[i]._NeruronsArray[j]._PB := _DownN * _NeruronsLevelArray[i]._NeruronsArray[j]._PB; end ; if _NeruronsLevelArray[i]._NeruronsArray[j]._PB>_MaxN then _NeruronsLevelArray[i]._NeruronsArray[j]._PB:=_MaxN; if PB > 0 then begin B := -_NeruronsLevelArray[i]._NeruronsArray[j]._PB; end else if (PB < 0 ) then begin B := _NeruronsLevelArray[i]._NeruronsArray[j]._PB; end else begin B := 0 ; end ; _NeruronsLevelArray[i]._NeruronsArray[j]._B := _NeruronsLevelArray[i]._NeruronsArray[j]._B + B; //-------------------------------- if I < (_LevelCount - 1 ) then begin ACount := _NeruronsLevelArray[i + 1 ]._NeruronsCount; for K := 0 to ACount - 1 do begin PW := _WeightMatrix[i].Matrix[j, k]._SumOffset; OldPw := _WeightMatrix[i].Matrix[j, k]._OldSumOffset; T := PW * OldPw; if T > 0 then begin _WeightMatrix[i].Matrix[j, k]._PW := _UpN * _WeightMatrix[i].Matrix[j, k]._PW; end else if (T < 0 ) then begin _WeightMatrix[i].Matrix[j, k]._PW := _DownN * _WeightMatrix[i].Matrix[j, k]._PW; end ; if _WeightMatrix[i].Matrix[j, k]._PW >_MaxN then _WeightMatrix[i].Matrix[j, k]._PW:=_Maxn; if PW > 0 then begin W := -_WeightMatrix[i].Matrix[j, k]._PW; end else if (PW < 0 ) then begin W := _WeightMatrix[i].Matrix[j, k]._PW; end else begin W := 0 ; end ; _WeightMatrix[i].Matrix[j, k]._Weight := _WeightMatrix[i].Matrix[j, k]._Weight + w; end ; end ; end ; end ; end ; constructor ObjNeuronsNet . create( const LevelArray: array of Integer ); begin InitNet(LevelArray); end ; destructor ObjNeuronsNet . Destroy; begin FreeNet; //hoho inherited ; end ; procedure ObjNeuronsNet . FreeNet; begin Self . _WeightMatrix := nil ; Self . _NeruronsLevelArray := nil ; Self . _SampleArray := nil ; end ; procedure ObjNeuronsNet . InitNet( const LevelArray: array of Integer ); var i, j: Integer ; begin _LevelCount := High(LevelArray) + 1 ; SetLength(_NeruronsLevelArray, _LevelCount); // Mem of Nerurons to Level SetLength(_WeightMatrix, _LevelCount - 1 ); //mem SetLength(_ExportArray, LevelArray[_LevelCount - 1 ]); Randomize; for i := 0 to _LevelCount - 1 do begin _NeruronsLevelArray[i]._NeruronsCount := LevelArray[i]; SetLength(_NeruronsLevelArray[i]._NeruronsArray, _NeruronsLevelArray[i]._NeruronsCount); //Mem if i < (_LevelCount - 1 ) then begin SetLength(_WeightMatrix[i].Matrix, LevelArray[i], LevelArray[i + 1 ]); //Hoho end ; for j := 0 to _NeruronsLevelArray[i]._NeruronsCount - 1 do begin if i = 0 then begin _NeruronsLevelArray[i]._NeruronsArray[j]._Type := 0 ; end else begin _NeruronsLevelArray[i]._NeruronsArray[j]._Type := 1 ; end ; _NeruronsLevelArray[i]._NeruronsArray[j]._PB := 0.1 ; _NeruronsLevelArray[i]._NeruronsArray[j]._B := Random * 2 - 1 ; //+/-1 end ; end ; //Init Neurons InitWeightMatrix; _StudyCount := 0 ; end ; procedure ObjNeuronsNet . InitWeightMatrix; //init Matrix Random var i, j, k: Integer ; FCount: Integer ; ACount: Integer ; F: Single ; begin Randomize; for i := 0 to _LevelCount - 2 do begin Fcount := _NeruronsLevelArray[i]._NeruronsCount; ACount := _NeruronsLevelArray[i + 1 ]._NeruronsCount; for j := 0 to Fcount - 1 do begin for K := 0 to ACount - 1 do begin _WeightMatrix[i].Matrix[j, k]._Weight := 2 * Random * F - F; _WeightMatrix[i].Matrix[j, k]._Offset := 0 ; _WeightMatrix[i].Matrix[j, k]._SumOffset := 0 ; _WeightMatrix[i].Matrix[j, k]._OldSumOffset := 0 ; _WeightMatrix[i].Matrix[j, k]._PW := 0.1 ; end ; end ; end ; end ; procedure ObjNeuronsNet . MemSample; var i: Integer ; begin SetLength(_SampleArray, _SampleCount); for i := 0 to _SampleCount - 1 do begin SetLength(_SampleArray[i].X, _NeruronsLevelArray[ 0 ]._NeruronsCount); SetLength(_SampleArray[i].Y, _NeruronsLevelArray[_LevelCount - 1 ]._NeruronsCount); end ; end ; procedure ObjNeuronsNet . SaveOffset; var i, j, k: Integer ; FCount: Integer ; ACount: Integer ; begin for i := 0 to _LevelCount - 1 do begin Fcount := _NeruronsLevelArray[i]._NeruronsCount; for j := 0 to Fcount - 1 do begin _NeruronsLevelArray[i]._NeruronsArray[j]._OldSumOffset := _NeruronsLevelArray[i]._NeruronsArray[j]._SumOffset; _NeruronsLevelArray[i]._NeruronsArray[j]._SumOffset := 0 ; if i < (_LevelCount - 1 ) then begin ACount := _NeruronsLevelArray[i+ 1 ]._NeruronsCount; for K := 0 to ACount - 1 do begin _WeightMatrix[i].Matrix[j, k]._OldSumOffset := _WeightMatrix[i].Matrix[j, k]._SumOffset; _WeightMatrix[i].Matrix[j, k]._SumOffset := 0 ; end ; end ; end ; end ; end ; procedure ObjNeuronsNet . Study; var i: Integer ; SumErr: Single ; begin if _StudyCount > _MaxStudyCount then Exit; SumErr := 0 ; for i := 0 to _SampleCount - 1 do begin CalcExport; //正向计算 CalcExportReturn; //反向 AddOffset; //累计偏差 SumErr := SumErr + GetError(); end ; Form1 . mmo1 . Lines . Add(IntToStr(_StudyCount)+ ':' +FloatToStr(SumErr)) ; Form1 . img1 . Canvas . Pixels[ Integer (_StudyCount* 5 ),floor( 150 -SumErr* 100 )]:= 255 ; if (SumErr/self . _SampleCount ) <= _MaxError then Exit; Changle; //改变 SaveOffset; //存储 _StudyCount := _StudyCount + 1 ; Study(); //迭代 end ; function ObjNeuronsNet . GetError: Single ; var SumErr: Single ; i: Integer ; begin SumErr := 0 ; for i := 0 to _NeruronsLevelArray[_LevelCount - 1 ]._NeruronsCount - 1 do begin SumErr := Sumerr + Power((_ExportArray[i] - _NeruronsLevelArray[_LevelCount - 1 ]._NeruronsArray[i]._V), 2 ); end ; Result := SumErr / 2 ; end ; end . |
delphi lazarus opengl
网页操作自动化, 图像分析破解,游戏开发