I really like the interaction and flexibility, but I failed to generate a set of colors as nice as those provided by Color Brewer (http://colorbrewer2.org/). Sure, Color Brewer doesn't dynamically generate color schemes, but the set of colors it provides looks damn good in just about any context. I might use this new app in cases where I need more than 12 colors (Color Brewer's limit), but I don't think dynamic color scheme generation adds much. The JSON and CSS exports do seem very useful, however, and a big improvement over Color Brewer's interface.
One of the reasons I'll probably stick with ColorBrewer palettes for now is that I can easily find color schemes that are safe for people with colorblindness or for display in particular environments.
It's interesting they've started this médialab in Science Po, which is (as the name implies) a social sciences university [0] in Paris [1].
Looking at the team [2] the name of Bruno Latour as a director seems a bit out of place. After his controversial views on the social construction of science [3] I find it astonishing that he is put at the head of the department that is in charge of bridging the world of the social scientists with that of the quantitative methods and hard sciences.
I don't see Latour as a particularly bad choice for bridging those worlds. One of his main points is paying close empirical attention to both how scientists actually work, and how non-human things work. Unlike a lot of humanities theorists of science, he actually went into laboratories and wrote down what seem like pretty honest accounts of what he found there. He's fairly controversial among pure social constructionists partly as a result. For example, he had a dispute with David Bloor some years ago [1] over Latour's insistence on including nonhuman objects in his explanation of how science works (rather than treating science as a purely social process). More recently, he's been attacking the tendency, born of the legacy of critical theory, to treat intellectual activity as a series of de-maskings of something the scholar will always find to be "naive" [2].
I mean, the last thing we should want, as scientists, is to have blind faith in a quasi-religious, idealized view of how science works, rather than empirically investigating how it actually works. Plus he seems pretty interested lately in tying in some philosophical ideas with quantitative network-analysis methods: http://www.bruno-latour.fr/sites/default/files/123-WHOLE-PAR...
This probably isn't the place to have a debate on Latour, but let me just say that it is possible to read him unsympathetically (as Sokal does) or sympathetically, to try to understand the viewpoint he is trying to get across. I've found it worthwhile to do the latter, but YMMV.
I more or less agree. I used to have a more negative view, but I realized that what you get from Sokal et al is a fairly inaccurate and biased reading.
Actually, while I thought the Sokal hoax itself was brilliant, I'm in retrospect quite disappointed by his follow-up book. It reads like a bit of a lazy hatchet-job by someone trying to put together an "attack" piece without having read the stuff he's attacking, and certainly not trying to read it sympathetically.
For the critique from the other end of the social constructionism spectrum, I strongly suggest you read the chapter devoted to him in Sokal and Bricmont's Impostures intellectuelles (I think it was translated as Fashionable nonsense).
It's a little hard to find in print, even in libraries. But I'm sure one could source a digital copy from a less reputable source.
I've used something similar in matlab, which I though worked a little better. It started out with red green blue, colors frequently considered maximally distinct, at least on color monitors for the non-colorblind. Then it just found maximally distant points in the color cube. I found it worked nicely and you could also seed it with your own starting color(s).
I'm no color expert but I'm pretty sure color distance in RGB space is a pretty poor metric compared to other color models (at least some of them). Looking at an RGB color cube I see big splashes of near-identical colors while the entire spectrum of oranges is hard to even find.
Interesting point. Though, some of the colors in OPs preset examples seemed very similar, even with only 7 colors. For generating graphs, maximally distinct colors seems the primary goal, even if they compete for attention, perhaps because of it.
As someone who has put together figures for a scientific paper, I can safely say that the last thing I want are a wide-range of colors to use. Many people are color-blind, and many people print on black & white printers. So when I design the figures, I stick with pure colors. Pure red (255,0,0) or pure green (0,255,0) or pure blue (0,0,255). I suppose it would be smarter to design a set of color codes where even if someone were missing one of the three color receptors in their eye, they would still be able to discern each individual color as distinct. Perhaps (128,128,0), (128,0,128), (0,128,128).
This will sound snarky, but there's just too much that's wrong with this comment to let it stand. For people reading this: please don't do any the above.
Go here for good color scales and spare yourself the risk of getting it all wrong: http://colorbrewer2.org
If you're going to not use colorbrewer, please, either know what you're doing (and if you're reading this comment, you don't) you either stick with black and white, or you at least use orange and blue (orange and blue is a combination that's safe with most common types of color-blindness, and also tends to separate well in black-and-white conversions).
Just as importantly: don't try to guess what colors look like for colorblind people, either (and don't do it in RGB space like the original post suggested either! RGB is a device-centric coordinate system; please use something like Lab or Luv). The rightest, simplest way to simulate different types of color blindness is to use define 2D affine subspaces of Lab space, like here: http://www.inf.ufrgs.br/%7Eoliveira/pubs_files/CVD_Mass_Spri... (figure 2, in page 3)
Aw, so sorry to tell you but it doesn't work in the latest version of IE. I'm not even sure if you should bother trying to get it to work but at minimum just put up a message to IE users telling them to use another browser.
This is an interesting tool, it fails at 16 colors. Which is surprising. (it may be the gamut of my display of course but I would think 16 would be within its abilities). That is probably the most number of distinct colors I've needed in a graph. Works nicely for numbers 10 and below.
Those are nice, I've bookmarked the protovis ones. Re-running my experiment, I seem to get two nearly identical magentas or two nearly identical greens every time I re-roll. It is entirely possible its my monitor though.
Great work!
Is there already an API to use? It would be great to use it directly and credit this website for the algorithms used.
The dynamical palette would become very useful in that case.
Is it hard to automatically generate a large (say 100+) palette of colors of increasing intensity that go well together? Do you have pointers to any algorithms for this?
Recently I was looking for something like this for a visualization. I had to find 16 nice looking distinct colors, which turned out to be pretty hard. Thanks!
