single layer peceptron

using js i made (and i trained) a single layer perceptron. a perceptron is also the most simple neural network model there is i'm studing machine learning and deep learning and for better understending concepts and algorithms i'm writing those algoritms in javascript and i admit having fun with it😅

//				=== THE MODEL	===

//the labels of our output is 0 or 1
function sign(n){
			return (n > 0) ? 1 : 0;//Heaviside step function
		}
class Perceptron{
	
	constructor(){
		// array de length 2 
		this.weigth = new Array(2);
		this.lrnRate = 0.01 // learning Rate
		this.bias=0 /*the bias help us to determinate our output when 
		input are only zeros*/
		this.epoch 
		this.maxEpochs=800

		// -1 < weigth < 1
		for(let i = 0; i< this.weigth.length; i++){	
			this.weigth[i]= ((Math.random()<=0.5)? -1 : 1)*Math.random()
		}
	}
	//the train method 
	train(input, target){
			
			for(let i = 0; i< this.weigth.length; i++){
			let guess = this.guess(input)
			let error = target - guess 
				//correcting weigths
				this.weigth[i]+= error*input[i]*this.lrnRate;
				}
		}
    	//guess method 
	guess(inputs){// feedforward
			let sum = 0
			for(let i = 0; i< this.weigth.length; i++){
				//the weigred sum
				sum += this.weigth[i]*inputs[i]
			}	 
			sum+= this.bias // Σwx + bias 
			//outputo of perceptron
			let output = sign(sum)
			return output;	
		}	
	//trainnig our model by epochs 	
	trainning(data, labels){/*trainnig our model by epochs means  repeat 
	the train as many time we set to converge the wgths to desired ones*/	
			this.epoch=0
			let output= new Array(data.length)
			while(output!=labels){
				for(let i = 0; i< data.length; i++){
					//calling the train
					this.train(data[i], labels[i])
					output[i] = this.guess(data[i])
				}
			this.epoch++
		if(this.epoch==this.maxEpochs){ // I give max 400 epochs 
        //to converge else break the loop
					break;
				}
			}
		}				
}
//                                by Talisma	:) enjoy it!

see the python version of this code on PERCEPTRON.ipynb

what problems can a single perceptron can solve?

a simple perceptron can only solve linearly separable problems

• How to know whether the data is linearly separable?

usually we can just visualise the data if its a 1, 2 or even 3 dimention data and you know, it really not worth it cause single layer percetron also becomes an unviable (bad) model when there is a the data has too much params(ex data(x, y, z, k))

somes example using this perceptron model

//		           === examples with logic gates ===
//
let dataset = [	[0,0],
		[0,1], 
		[1,0],
		[1,1]]
//or gate
let OR = new Perceptron()
OR.bias=0, OR.lrnRate= 0.5
OR.trainning( dataset, [0,1,1,1])// 80 epochs nedded

end gate

let AND= new Perceptron()
OR.trainning( dataset, [0,1,1,1])// 80 epochs nedded
AND.bias = -1
AND.lrnRate = 0.05 
AND.trainning( dataset, [0,0,0,1]) // 400 epochs nedded

nand gate

let NAND = new Perceptron()
NAND.bias = 1, NAND.lrnRate= 0.5
NAND.trainning( dataset, [1,1,1,0]) 

try it for yourself here

🤩 special thanks for "the coding train" for their marvelous neural network playlist on youtube