www.algodoo.com

Posted: **Sat Mar 26, 2016 11:42 pm**

That scene illistrate principles of neural networks, evolution, mutation and selection.
Selecting best bugs and save-load it for researching you could get incredible fast ones, able to pass maze for less than 20 sec :thumbup:

Evoluton of bugs (neural network brain)

Kilinich

Rating:

Filesize: 293.48 kB
Comments: 17
Ratings: 8

Posted: **Sun Mar 27, 2016 1:07 am**

Kind of like BoxCar2D 8-)

So basically what it does is it generates a bug, and then it figures out what it did right and wrong, and applies all that it learns to the next bug in the sequence?

Posted: **Sun Mar 27, 2016 1:45 pm**

pnvv wrote:Kind of like BoxCar2D

So basically what it does is it generates a bug, and then it figures out what it did right and wrong, and applies all that it learns to the next bug in the sequence?

Yes, it has "stable" config with previous best result and update it on collision if new result is better, overwise - generate new sligtly different from last stable.

Posted: **Sun Mar 27, 2016 11:14 pm**

In your comment to me on Algobox you said, "..... key to success is selection rule.
So you could build anything like round track and choose criteria how to select better performance.... "

How do you choose criteria for better performance? Is there some way to change the selection rule? Please explain.

Posted: **Mon Mar 28, 2016 6:23 am**

Amazing work, I love evolutionary algorithms. It's a fantastic visualisation.

Posted: **Mon Mar 28, 2016 11:04 pm**

I got besttime of 23,
with this bestConfig:

Code: Select all: [[[[-1.2128481, -0.55738807, 0.64659858, -0.26619381, -1.1063217], [0.89303547, 1.5829785, 1.0322715, -0.16735084, -2.1852973], [-1.5183198, -0.52044833, -0.62832379, 0.082131788, -0.93296903], [1.5706787, 0.51257092, -0.098530829, -0.49385658, 0.85188723], [1.2936145, 0.16742918, 0.15303953, -1.4831454, -0.586002]], [0.29760474, 1.521318, -0.10459325, 0.59665453, 1.0448649]], [[[0.52260596, 0.45794782, 0.12360375, 0.70862603, 0.53481853], [1.0198236, -2.3892474, -0.65290409, -1.4095669, -1.2932711]], [-0.61260831, -0.97459435]]]

Do we have a chance to get enabled some manipulations of weights "by hand"?
i like it!! thanks again. :clap:

Posted: **Tue Mar 29, 2016 12:08 pm**

DrAgon wrote:I got besttime of 23,
Do we have a chance to get enabled some manipulations of weights "by hand"?
i like it!! thanks again.

Sure - you could edit evolution-save.cfg file or scene.my.bestConfig variable

cfg stucture: [config, topology]
config (also bug's _config or scne.my.bestConfig): [hiddenLayer, outputLayer]
hiddenLayer, outputLayer: [neuron1,neuron2... neuronN]
neuron: [[inputWeight1, inputWeight2... inputWeightN], treshHold]
topology: [inputsCount, hiddenNeuronsCount, outputNeuronsCount]