adjustments

src/genetic/methods/selection.ts

@@ -3,14 +3,14 @@ import { Population } from "../model/population";
 
 export class Selection {
 	static RANK = (selectedCandidatesNumber?: number) => <T>(candidates: Candidate<T>[]): Candidate<T>[] => {
-		const selectedCandidatesSize = selectedCandidatesNumber || Math.floor(candidates.length * 1);
+		const selectedCandidatesSize = selectedCandidatesNumber || Math.floor(candidates.length * 0.8);
 		/**
 		 * Sort candidates in ascending fitness order
 		 */
 		const sortedCandidates = candidates.sort((a, b) => a.fitness(a.genes) - b.fitness(b.genes));
 		const selectedCandidates: Candidate<T>[] = [];
 		const candidatesSize = sortedCandidates.length;
-		for (let i = 0; i < selectedCandidatesSize - 1; i++) {
+		for (let i = 0; i < selectedCandidatesSize; i++) {
 			const pDomain = candidatesSize * (candidatesSize + 1) / 2;
 			const p = Math.floor(Math.random() * pDomain);
 			let bornSup = 1;
@@ -22,10 +22,6 @@ export class Selection {
 				bornSup += j + 1;
 			}
 		}
-		/**
-		 * The best candidate cheats and always win
-		 */
-		selectedCandidates.push(sortedCandidates[candidatesSize - 1]);
 		return selectedCandidates;
 	}
 

src/genetic/model/candidate.ts

@@ -19,7 +19,7 @@ export class Candidate<T extends Gene> {
 		this.fitness = options.fitness;
 		this.mutate = options.mutate;
 		this.crossProbability = options.crossProbability || 0.8;
-		this.mutationProbability = options.mutationProbability || 0.1;
+		this.mutationProbability = options.mutationProbability || 0.2;
 	}
 
 	cross = (other: Candidate<T>) => {

src/genetic/model/population.ts

@@ -8,6 +8,7 @@ interface PopulationOptions<T> {
 	mutate?: (genes: T[]) => T[];
 	mutationProbability?: number;
 	crossProbability?: number;
+	ellitism?: boolean;
 }
 
 export class Population<T extends Gene> {
@@ -17,6 +18,7 @@ export class Population<T extends Gene> {
 	mutate: (genes: T[]) => T[];
 	mutationProbability?: number;
 	crossProbability?: number;
+	ellitism: boolean;
 
 	/**
 	 * Generate a new randomized population
@@ -43,6 +45,7 @@ export class Population<T extends Gene> {
 		this.select = options.select || Selection.RANK();
 		this.fitness = options.fitness || (() => 0);
 		this.mutate = options.mutate || (g => g);
+		this.ellitism = options.ellitism || false;
 	}
 
 	createNextGeneration = (): Population<T> => {
@@ -55,7 +58,6 @@ export class Population<T extends Gene> {
 		 * Cross stronger candidates and mutates their children
 		 */
 		const newCandidates = this.cross(selectedCandidates);
-
 		/**
 		 * Add the children to the population
 		 */
@@ -66,7 +68,14 @@ export class Population<T extends Gene> {
 		 */
 		const cleanedNewGeneration = newGeneration
 			.sort((a, b) => b.fitness(b.genes) - a.fitness(a.genes))
-			.slice(0, this.candidates.length);
+			.slice(0, this.candidates.length - +this.ellitism);
+
+		if (this.ellitism) {
+			// Keep the best oldest candidate
+			cleanedNewGeneration.push(
+				this.candidates.sort((a, b) => b.fitness(b.genes) - a.fitness(a.genes))[this.candidates.length - 1]
+			);
+		}
 
 		return new Population(cleanedNewGeneration, this.getOptions());
 	}

src/index.ts

@@ -1,56 +1,62 @@
 import { Selection } from "./genetic/methods";
 import { Candidate, Population } from "./genetic/model";
-import { Network } from "./neural-network/network";
-
-const fitness = (nn: Network) => () => {
-	// console.log("FITNESS");
-	// const xorFF = nn.activate([0, 0])[0];
-	// const errFF = 1 / (1 - xorFF);
-	// // console.log(xorFF);
-	const xorFT = nn.activate([0, 1])[0];
-	// const errFT = 1 / xorFT;
-	// // console.log(xorFT);
-	// const xorTF = nn.activate([1, 0])[0];
-	// const errTF = 1 / xorFT;
-	// // console.log(xorTF);
-	// const xorTT = nn.activate([1, 1])[0];
-	// const errTT = 1 / (1 - xorFF);
-	// // console.log(xorTT);
-	// // console.log(1 - xorFF + xorFT + xorTF + 1 - xorTT);
-	// console.log(errFF, errFT, errTF, errTT);
-	// return Math.pow(1 / (errFF + errFT + errTF + errTT), 2);
-	return xorFT;
+
+const PASSWORD = "hardcorepasswordmasterrace";
+const fitness = (password: string) => (genes: string[]) => {
+	const passwordArray = password.split("");
+	if (genes.length !== password.length) {
+		throw Error("passwords length not matching");
+	}
+	let score = 0;
+	for (let i = 0; i < genes.length; i++) {
+		if (genes[i] === passwordArray[i]) {
+			score += 1;
+		}
+	}
+	return score * 100 / password.length;
 };
 
-function mutate(weights: number[]) {
-	const weightsCopy = weights.slice();
-	const indexMutated = Math.floor(Math.random() * weightsCopy.length);
-	weightsCopy[indexMutated] = Math.random() * 2 - 1;
-	return weightsCopy;
-}
+const mutate = (password: string) => (genes: string[]) => {
+	const genesCopy = genes.slice();
+	const mutatedIndex = Math.floor(Math.random() * PASSWORD.length);
+	genesCopy[mutatedIndex] = getRandomLetter();
+	return genesCopy;
+};
 
 // Create the networks
 const candidates = [];
 
-for (let i = 0; i < 20; i ++) {
-	const nn = Network.perceptron([2, 4, 3, 1]);
-	const candidate = new Candidate(nn.weights, {
-		fitness: fitness(nn),
-		mutate,
-		mutationProbability: 1
-	});
+for (let i = 0; i < 100; i ++) {
+	const genes = generateRandomGenes();
+	const candidate = new Candidate<string>(
+		generateRandomGenes(),
+		{
+			fitness: fitness(PASSWORD),
+			mutate: mutate(PASSWORD),
+			mutationProbability: 0.4
+		}
+	);
 	candidates.push(candidate);
 }
 
 const initialPop = new Population(candidates);
 
 let pop = initialPop;
 
-for (let i = 0; i < 3; i ++) {
-	if (true) {
-		console.table(pop.candidates.map(it => it.fitness([])));
-	}
+for (let i = 0; i < 100; i ++) {
 	document.getElementById("main")!.innerHTML
-		+= ("<br/><br />" + pop.candidates[0].fitness([]));
+		+= ("<br/><br />" + pop.candidates[0].genes + ", fitness: " + pop.candidates[0].fitness(pop.candidates[0].genes));
 	pop = pop.createNextGeneration();
 }
+
+function generateRandomGenes() {
+	const genes = [];
+	for (let i = 0; i < PASSWORD.length; i++) {
+		genes.push(getRandomLetter());
+	}
+	return genes;
+}
+
+function getRandomLetter() {
+	return String.fromCharCode(97 + Math.floor(Math.random() * 26));
+}

tslint.json

@@ -21,6 +21,7 @@
 		"arrow-parens": false,
 		"array-type": [true, "array"],
 		"variable-name": [true, "check-format", "allow-leading-underscore"],
-		"max-line-length": [true, 130]
+		"max-line-length": [true, 130],
+		"prefer-for-of": false
 	}
 }