Improve RGB mode

[kaijagahm]

Added a pretty primitive version of [RGB mode] in https://github.com/kaijagahm/ygdpDashboard/commit/94d9ec13802d164af150bf195b53689a46c8e6a3.

I don't love this, though--I don't think it's a good way to display data. Problems:

• Colors get muddy because they go to black. I can't figure out a way to change this: the rgb() function takes a max argument but not a min argument. I can't figure out if there are any other "triangular" functions like rgb() that can take arbitrary colors.
• Red/green colorblind people won't be able to read this map. That's a huge flaw. And as mentioned above, I can't figure out a way to change the colors to something other than r/g/b.
• The legends in leaflet::addLegend() are created as svg objects, according to https://community.rstudio.com/t/increase-decrease-size-of-leaflet-legend/17037. Therefore, you can't really change the size. That's why there are three different legends in three different corners, which looks really awkward.

Going to bring these suggestions for tweaks into a different issue.

Originally posted by @kaijagahm in https://github.com/kaijagahm/ygdpDashboard/issues/25#issuecomment-746828757

[kaijagahm]

A different idea that Jim and I had was to simply replace the three separate color legends with a single triangle legend, like this:

(Note: the above image is a screenshot from https://observablehq.com/@toja/triangular-color-interpolation, which is a very useful website for computing color interpolations. Could definitely be helpful for the future, if we want to develop a not-horrible color triad.)

So basically, we were thinking I could just take a screenshot from that site, label the vertices "Sentence 1", "Sentence 2", an "Sentence 3", overlay that static image on the map instead of all of the legends, and call it good.

But... that's not good enough. Because we also have black and white going on, somehow. The sticking point is that I'm generating the colors through the rgb() function within R, and I don't actually understand the mechanics behind it. For example, take a look at this screenshot from an example with three "after perfect" sentences in interpolation mode:

Some of the hexagons are almost white, and others are almost black! What's going on here?

I think the triangle actually isn't a proper representation of the colorspace. It's not clear to me how you'd get to the internal colors in the triangle. After all, the colors on the map are generated by manipulating the three components (R, G, and B) separately. If all of them are set to 255, then the color is white (so, by analogy, if all three sentences are rated 5 in a given hexagon, then that hexagon would be white). If all of them are set to 0 (i.e. if all three sentences are rated as 1), then that hexagon would be black. I don't actually think a triangle represents what's going on here. Maybe we need a triangular prism?? But no, even that doesn't seem right, because I still really don't understand how you would reach the center of the triangle.

Aha! I did some more googling, and it seems that the correct representation of the colorspace is as a cube. There's a good simulator for this here: https://programmingdesignsystems.com/color/color-models-and-color-spaces/index.html#:~:text=The%20RGB%20color%20model%20is,the%20bounds%20of%20the%20cube

That website allows you to input different colors and look at the results. If you put all sliders up to 100%, you get white; if you put them all down to 0%, you get black.

The problem is, of course, that this is probably not possible to display as a legend on a graph. Well, there you go. This explains why RGB is never really used for data viz like this.

[kaijagahm]

After discussion with Jim on 3/11 and 3/12, changed the max color to 130 out of 255 for each of R, G, and B. This makes the map a little more readable.

Made this change in commit https://github.com/kaijagahm/ygdpDashboard/commit/46ddb123f5618ae075e8d012fde8c20abc1a29fd

[kaijagahm]

This thesis talks a bit about GabMap's approach to using three colors on an MDS map (figure at the end refers to section 6.something.something, which describes a bit.

The main difference between their approach and ours seems to be that they plotted the results of an MDS analysis, whereas we're trying to plot just the raw IDW data. They did MDS and then IDW. Not sure what the difference is?

Measuring_Dialect_Pronunciation_Di_erenc.pdf