Back Propagation Neural Network ( BPN ) is one of neural-network algorithms.
#import "KRBPN.h"
@interface ViewController ()
@property (nonatomic, strong) KRBPN *_krBPN;
@end
@implementation ViewController
@synthesize _krBPN;
- (void)viewDidLoad
{
[super viewDidLoad];
/*
* @ 本參數設定約耗去 200 KB Memory
*/
_krBPN = [KRBPN sharedNetwork];
//各輸入向量陣列值
_krBPN.inputs = [NSMutableArray arrayWithObjects:
//X1
@[@1, @2, @0.5, @1.2],
//X2
@[@0, @1, @0.3, @-0.9],
//X3
@[@1, @-3, @-1, @0.4],
nil];
/*
* @ 輸入層、隱藏層、輸出層之間的神經元初始權重
*
* - W14 : 輸入層 X1 到隱藏層 Net 4
* - W15 : 輸入層 X1 到隱藏層 Net 5
*
* - W24 : 輸入層 X2 到隱藏層 Net 4
* - W25 : 輸入層 X2 到隱藏層 Net 5
*
* - W34 : 輸入層 X3 到隱藏層 Net 4
* - W35 : 輸入層 X3 到隱藏層 Net 5
*
* - W46 : 隱藏層 Net 4 到輸出層的 Net 6
* - W56 : 隱藏層 Net 5 到輸出層的 Net 6
*
*/
//輸入層各向量值到隱藏層神經元的權重 ( 連結同一個 Net 的就一組一組分開，有幾個 Hidden Net 就會有幾組 )
_krBPN.inputWeights = [NSMutableArray arrayWithObjects:
//W14, W15
@[@0.2, @-0.3],
//W24, W25
@[@0.4, @0.1],
//W34, W35
@[@-0.5, @0.2],
nil];
//隱藏層神經元的偏權值
_krBPN.hiddenBiases = [NSMutableArray arrayWithObjects:
//Net 4
@-0.4,
//Net 5
@0.2,
nil];
//隱藏層神經元到輸出層神經元的權重值
_krBPN.hiddenWeights = [NSMutableArray arrayWithObjects:
//W46
@-0.3,
//W56
@-0.2,
nil];
//有幾顆隱藏層的神經元 ( 不用外部設定，由偏權值數目自動設定 )
//_krBPN.countHiddens;
//輸出層神經元偏權值, Net 6 for output
_krBPN.outputBias = 0.1f;
//期望值
_krBPN.targetValue = 1.0f;
//學習速率
_krBPN.learningRate = 0.8f;
//收斂誤差值 ( 一般是 10^-3 或 10^-6 )
_krBPN.convergenceError = 0.001f;
__block typeof(_krBPN) _weakKrBPN = _krBPN;
//每一次的迭代( Every generation-training )
[_krBPN setEachGeneration:^(NSInteger times, NSDictionary *trainedInfo){
NSLog(@"Generation times : %i", times);
//NSLog(@"trainedInfo : %@\n\n\n", trainedInfo);
}];
//訓練完成時( Training complete )
[_krBPN setTrainingCompletion:^(BOOL success, NSDictionary *trainedInfo, NSInteger totalTimes) {
if( success )
{
if( !_weakKrBPN.trainedNetwork )
{
[_weakKrBPN saveTrainedNetwork];
}
NSLog(@"Training done with total times : %i", totalTimes);
NSLog(@"TrainedInfo : %@", trainedInfo);
NSLog(@"TrainedNetwork with inputWeights : %@\n\n\n", [_weakKrBPN.trainedNetwork.inputWeights description]);
}
}];
//Remove your testing trained-network records.
[_krBPN removeTrainedNetwork];
//Start the training network, and it won't be saving the trained-network when finished.
[_krBPN training];
//Start the training network, and it will auto-saving the trained-network when finished.
[_krBPN trainingDoneSave];
//If you wanna pause the training.
[_krBPN pause];
//If you wanna continue the paused training.
[_krBPN continueTraining];
//If you wanna reset the network back to initial situation.
[_krBPN reset];
//When the training finished, to save the trained-network into NSUserDefaults.
[_krBPN saveTrainedNetwork];
//If you wanna recover the trained-network data.
[_krBPN recoverTrainedNetwork];
//Or you wanna use the KRBPNTrainedNetwork object to recover the training data.
KRBPNTrainedNetwork *_trainedNetwork = [[KRBPNTrainedNetwork alloc] init];
_trainedNetwork.inputs = [NSMutableArray arrayWithObjects:
@[@1],
@[@0],
@[@1],
nil];
[_krBPN recoverTrainedNetwork:_trainedNetwork];
//To remove the saved trained-network.
[_krBPN removeTrainedNetwork];
}
@endV1.0
MIT.