VURD to the third power...

[triadruid]

I would like to graph several pairs of RGB triplets (#FFBB99,#000020 for example), but cannot figure out how to do it properly without folding space and time in unfortunate ways.

Any ideas that don't involve six dimensions?

Comments

[featherynscale.livejournal.com]

Why six dimensions instead of three?

Also, why are you doing this at all?

[triadruid.livejournal.com]

Pairs of colors: RGB for foreground, RGB for background. I might be able do line segments connecting the two points in a 3D-graph...

It's mostly just idle speculation. I've been staring at computer guts too long today. :)

[auraseer.livejournal.com]

Yeah, line segments was my first thought.

Or if you want to get trippy, plot the first triplet as a point, and then draw the point in the color indicated by the second triplet. It would be tough to interpret but might generate pretty pictures.

[triadruid.livejournal.com]

Your second suggestion actually might be kind of interesting.

[rfunk.livejournal.com]

Why not treat each triplet as a single number? Then you're down to two dimensions, and you just have to scale them down a lot.

[auraseer.livejournal.com]

That makes it hard to see anything useful by looking at the distance between points.

For instance, #000001 is pretty much the same color as #010000; they are nearly black and hard to distinguish from each other. On a 3D graph, that is easy to see because they are both close to the origin. However if you flatten them into single numbers and stick them on a number line, the red one will be thousands of times farther from the origin than the blue one.

[triadruid.livejournal.com]

That was my first thought, but FF,00,99 is a very different shade than FF,01,00.

er...

[popefelix.livejournal.com]

Well, those are just integers, so why not graph them that way? Maybe on a logarithmic scale so that you don't have vast amounts of empty space in between your points.

Re: er...

[triadruid.livejournal.com]

*scratches head* I'm not sure I explained myself well originally. But the integers are linked because they describe the color along three axes, so unless I'm misunderstanding something...

Re: er...

[popefelix.livejournal.com]

let's step back a moment. What are you trying to accomplish?

Re: er...

[triadruid.livejournal.com]

It's actually sort of a silly exercise, but I wanted to identify where my LJ Friends 'custom color scheme' was not distinguishing well between people. My flist has gotten big enough lately that I've started to double-up on color combinations, and when somebody new gets the same basic color scheme as somebody I've been reading for a long time (dark blue on peacock blue, for example), it throws my brain off.

So I thought, if I could show them graphically rather than the laundry list of http://www.livejournal.com/friends/edit.bml, it would be systematic. But then my brain went splodey. :)

Re: er...

[popefelix.livejournal.com]

YOUR HEAD ASPLODE! ;)

Sounds interesting. But I don't think the bandwidth is there to do it in LJ comments. Let's talk about it tomorrow night over pool.

[auraseer.livejournal.com]

You could graph each pair as a line segment. If they're ordered pairs, graph them as vectors (with the arrow on one end) or just give them different-colored endpoints.

More generally, a set of points can be represented as a connected shape. Three points can be drawn as a triangle, four as a quadrilateral, etc. For flat triangles or complete solids, shading or coloring the interior makes them easier to distinguish from each other.

Why are you trying to graph RGB triplets instead of just drawing the colors?

[triadruid.livejournal.com]

Drawing the colors might work, but I'm looking for insufficient contrast between pairs, and my eyes don't always do a very good job of color matching. Also, there's more pairs than I can put side-by-side here.

I think the line segments are going to end up being the way to go here.

[auraseer.livejournal.com]

If all you want to know is the contrast, you don't need to generate a graph. Just use the distance formula to calculate the separation of each pair of points. That gives you a single number for each pair, and a larger number indicates greater contrast.

Given points A (Xa, Ya, Za) and B (Xb, Yb, Zb), plug the values into this:

sqrt( (Xa-Xb)2 + (Ya-Yb)2 + (Za-Zb)2 )

[triadruid.livejournal.com]

Mmmm, problem is, I need to compare all of them pairwise, not just discreet pairs. (see my response to [Unknown site tag] above).

It'd still only be 27k pairs or so, but... ;)

[auraseer.livejournal.com]

You want to compare each value to every other? I don't see how you'll do that with the line segment graph. It will generate a big impenetrable ball of tangles.

One line segment represents one ordered pair, so if there are 27k possible pairings, you'll wind up with 27k line segments crammed into the graph. With all those overlapping and obscuring each other, I have no idea how you'd read it.

[triadruid.livejournal.com]

Nah, I should only end up with √27,000 line segments - which might still be impenetrable. :)

[geekatlarge.livejournal.com]

What did you intend to graph these in? Also, I'd like to see the full set of points if possible.