Awesome

About

KRDelta is a supervised learning which Delta rule of Machine Learning, it still works in recommendation system and real-time user behavior analysis on mobile apps.

Podfile

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

How To Get Started

Import

#import "KRDelta.h"

Normal Case

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);
}];

Saving & Fetching Trained Neuron

If neuron has finished training that we could save it through KRDeltaFetcher in completion block or anywhere.

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

Setting Weights by Yourself

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

Version

V1.6.0

LICENSE

MIT.