Galapagos graphical feedback question

Hi,

I was wondering if there has been posted information about the graphical feedback from the Galapagos component. I'm really impressing people around here when I run these but it'd be even more impressive if I knew what I was seeing... :ø)

This one is the most obvious to me:

In this case, there are 4 genes (X-axis) and the slider positions are along the Y-axis. But which gene is which input slider? The appear to be populating the X-axis randomly and shift position between runs...?

Not so sure about this one:

OK, the number is the fitness of the individual and the bar to the right of it has 4 parts so that would represent the 4 genes in this model. But what about the colors and the order in which they appear?

No clue about this one:

... except that the 12 dots represent the top 25% individuals in the current generation. But why are there sometimes red crosses and how does the position relate to the fitness?

As for the main display:

I understand that higher is better but what about the colors and the thick line?

Thanks for a fantastic program!

cheers,

wim

Replies to This Discussion

Permalink Reply by David Rutten on November 26, 2010 at 1:37pm

Hi Wim,

1) The order of the sliders should be the same for every run. The Galapagos object Genome input maintains a list of Slider IDs and this list is only altered when you create new wires or remove existing wires.

2) The capsule display is very similar to the first graph, but instead of drawing a line connecting relative y-values for each slider, each slider get's assigned a colour (from dark red to yellow) based on it's relative position. It allows you to see whether two genomes are similar or not without taking up too many y pixels.

3) This is a tricky one to explain. Every genome in a single species has the same 'dimensionality'. For example, if there are only two sliders you can say that the entire genome space for the species is 2-dimensional. For every possible combination of these two sliders, there is a fitness value (or a height) on this two dimensional plane. If your genome consists of 6 sliders, then we're talking about a 6-dimensional space.
As you probably know, distances between points are computed with the same formula, regardless of the dimensions of these points. Pythagoras' method works for all points with identical and integer dimensions. So even though I cannot display a 6-dimensional genome space on a two-dimensional computer screen, I can compute the distances between all the genomes in a species/generation. This then gives me a matrix with the distances from every genome to every other genome. I translate this distance matrix to a node-spring particle system and solve that system in two-dimensions, which ultimately results in the point-scatter graph you see on the screen.
The axes of this 2D representation of the ND distances are meaningless. The absolute position of the points inside this grid are governed partly by chance. However the relative positions are meaningful in that they convey which genomes are similar and which ones are different. Points which appear close together represent similar genomes, points which appear far apart represent different genomes.
Basically it becomes very simple to see the entire collection of genomes and get a feel for how varied the set is. You can often even see sub-species appear as distinct clusters of points.

4) For every generation, I display the fittest genome (upper boundary of yellow area), the worst genome (lower boundary of yellow area), average genome fitness (the thick red line) and the standard deviation of the fitness distribution in both directions (the orange area). Everything below the average is hatched.

Have you seen the Blog entry about galapagos?

--
David Rutten
david@mcneel.com
Seattle, WA

Permalink Reply by Wim R.M. Dekeyser on November 29, 2010 at 10:58pm

Thanks for this answer, David!
I have now also read the blog entry and it's all pretty clear, thanks!
.. but how about the red crosses that appear in the genome distance plot (number 3 in your answer)?

cheers,
wim

RSS