KRDelta 1.6.0

KRDelta 1.6.0

×

测试已测试✗
Lang语言 Obj-CObjective C
许可证 MIT
发布最后发布2017年6月

由 Kalvar Lin 维护。

KRDelta 1.6.0

关于

KRDelta 是一种监督学习，它采用Delta规则进行机器学习，它仍然应用于移动应用程序中的推荐系统和实时用户行为分析。

Podfile

platform :ios, '8.0'
pod 'KRDelta', '~> 1.6.0'

如何开始

导入

#import "KRDelta.h"

常规情况

KRDelta *delta         	  = [KRDelta sharedDelta];
delta.activeFunction   	  = KRDeltaActivationTanh;
delta.optimization.method = KRDeltaOptimizationFixedInertia;
delta.batchSize           = 1; // kKRDeltaFullBatch or 0 is full-batch, 1 is standard SGD, > 1 is mini-batch.
delta.learningRate     	  = 0.8f;
delta.convergenceValue 	  = 0.001f;
delta.maxIteration     	  = 100;
[delta addPatterns:@[@1.0f, @-2.0f, @0.0f, @-1.0f] target:-1.0f];
[delta addPatterns:@[@0.0f, @1.5f, @-0.5f, @-1.0f] target:1.0f];
[delta addPatterns:@[@1.0f, @1.0f, @1.5f, @0.5f] target:1.0f];
[delta addPatterns:@[@2.0f, @0.5f, @-1.5f, @-0.5f] target:1.0f];
[delta addPatterns:@[@-0.5f, @-1.5f, @0.5f, @1.0f] target:-1.0f];
[delta setupRandomMin:-0.5f max:0.5f];
[delta randomWeights];
[delta trainingWithIteration:^(NSInteger iteration, NSArray *weights) {
    NSLog(@"Doing %li iteration : %@", iteration, weights);
} completion:^(BOOL success, NSArray *weights, NSInteger totalIteration) {
    NSLog(@"Done %li iteration : %@", totalIteration, weights);
}];

保存和获取已训练的神经元

如果神经元完成训练，我们可以在完成块或其他地方通过 KRDeltaFetcher 来保存它。

// Saving
[delta trainingWithCompletion:^(BOOL success, NSArray *weights, NSInteger totalIteration) {
	[[KRDeltaFetcher sharedFetcher] save:delta forKey:@"A1"];    
}];
// Fetching
KRDelta *trainedDelta = [[KRDeltaFetcher sharedFetcher] objectForKey:@"A1"];

自行设置权重

[delta setupWeights:@[@1.0f, @-1.0f, @0.0f, @0.5f]];

批量学习

[delta setBeforeUpdate:^BOOL(NSInteger iteration, NSArray *deltaWeights){
    // YES: keep training.
    // NO:  stop continue.
    return YES;
}];

版本

V1.6.0

LICENSE

MIT.