ScratchData LogoScratchData
Back to cactus3's profile

Genetic Algorithm training neural network!

CAcactus3•Created April 29, 2017
Genetic Algorithm training neural network!
12
10
543 views
View on Scratch

Instructions

this project uses a genetic algorithm to find the all 17 weights of this neural network to map 2 inputs to one output. green circles are biases and always have a weight of 1

Description

How the genetic algorithm works (step by step): 1. create 10 sets of 17 random weights 2. run them all through the neural network and select the 2 best sets that have the lowest error (highest fitness) 3.delete all of the old sets and create 10 new sets based on the 2 best sets in step 2. 4. all of the 10 new sets have a 1/10 chance of mutating (having 4 of 17 values set to random again) How the neural network works: watch this video: https://youtu.be/bxe2T-V8XRs?t=2m31s

Project Details

Project ID158036259
CreatedApril 29, 2017
Last ModifiedOctober 14, 2017
SharedApril 29, 2017
Visibilityvisible
CommentsAllowed