These are chat archives for harthur/brain

30th
Dec 2014
Heather Arthur
@harthur
Dec 30 2014 02:21
@l33z3r The output from a brain.js neural network is not binary, and is an array of continuous values from 0 to 1. I see for your input you are using discrete values (like 2, 3, etc.), those need to be continuous as well, scaled from 0 to 1 as an example (like 0.6).
Heather Arthur
@harthur
Dec 30 2014 02:28
What's the problem you're trying to solve?
Lee Farrell
@l33z3r
Dec 30 2014 09:17
I was just playing around with the libraby trying to get a better understanding. I wanted to simply model a linear function y = 2*x
but I think the problem in my example is I am using integers that are not scaled!
thanks for the reply, great work on the library!
I am going to use it to run an AI air hockey agent.
written with box2d
If you have any advice on this, any help would be appreciated!
Lee Farrell
@l33z3r
Dec 30 2014 09:26
Actually, when I use the following training data, the network still will not learn the mapping:
net.train([{input: [.01], output: [.02]},
{input: [.02], output: [.04]},
{input: [.03], output: [.06]},
{input: [.04], output: [.08]}]);