Max gamut

[Bisaloo]

fix #66

This is something that I have been wanting to do for a long time and that I now have to do for a manuscipt in revision: I need to compute the maximum theoretical gamut given a visual system and illuminant.

This is a first pass at trying to do this but I'm not entirely sure it's the best way. I also really need to refactorize dispace()/trispace()/tcspace() at some point (plus address #102 and #179 while I'm at it) but I don't know if I will manage to do it in this PR.

An example of what it produces with triplot():

[thomased]

Related #66 — would be awesome.

[Bisaloo]

Quality check: cie monochromatic loci

data(sicalis)
vis_sicalis <- vismodel(sicalis, visual = "cie10", illum = "D65")
tri_sicalis <- trispace(vis_sicalis)
plot(tri_sicalis, gamut = TRUE)

[Bisaloo]

Currently stuck at: vismodel() creates -Inf values for qcatch = "fi". I need to figure out what we're supposed to do with Fechner's law when Qi = 0.

[thomased]

This looks awesome.

Yeah the Qi = 0 and Qi < 1 cases for log-transforming are longstanding. I don't have a clean answer, no-one seems to mention it which has made me think I'm missing something obvious...

[thomased]

This is super cool. Couple of quick Q's while I'm playing around:

• (1) Points should always fall inside the gamut right? Unless I'm misunderstanding how it works. e.g. centre
  
• (2) It's not currently implemented for qcatch = 'fi' just yet correct? Because of the above Qi = 0 issue.

EDIT: Right, I see 687f18c, ignore (2)

[Bisaloo]

No, you're right, points should not fall outside. It's likely a bug. Could you share the code you used to produce these graphs please?

On a related note, the first graph is strange because the max gamut should always be convex and it's not the case here.

[thomased]

Lazy code:

data(flowers)

vis_a <- sensmodel(c(350, 450, 550))
names(vis_a) <- c('wl', 's', 'm', 'l')
mod_a <- vismodel(flowers, vis_a)
tri_a <- colspace(mod_a)

vis_b <- sensmodel(c(340, 530, 550))
names(vis_b) <- c('wl', 's', 'm', 'l')
mod_b <- vismodel(flowers, vis_b)
tri_b <- colspace(mod_b)

vis_c <- sensmodel(c(330, 340, 660))
names(vis_c) <- c('wl', 's', 'm', 'l')
mod_c <- vismodel(flowers, vis_c)
tri_c <- colspace(mod_c)

par(mfrow = c(1, 3))
plot(tri_a, gamut = TRUE)
plot(tri_b, gamut = TRUE)
plot(tri_c, gamut = TRUE)

[Bisaloo]

Random thought while trying to debug this: should we provide a method for identify() for colspace objects?

[Bisaloo]

Okay, the gamut for trispace objects is now correct. The bug was not present for dispace and tcspace objects.

On a related note, what is the best way to figure the gamut in plots?

[Bisaloo]

Issue: conversion to relative qcatches when all qcatches are zero (e.g. CIE between 300 and 390 nm) creates NaN.

[thomased]

This is looking awesome. Couple of quick things:

(1) Uninformative error atm when trying to use transformed qcatches

data(flowers)
vis.flowers <- vismodel(flowers, qcatch = 'Ei', visual = "apis", scale = 10000)
tri.flowers <- colspace(vis.flowers)
plot(tri.flowers, gamut = TRUE)

Error in convhulln(coords_mono) : 
The first argument should not contain any NAs 

(2) Not sure what your thoughts are, but if it'll be a while before before we solve/implement it with transformed qcatches it's probably worth mentioning that in the docs (...currently only possible when qcatch = Qi or something).

(3) For tetrachromats, the volume when plotted looks like it's basically ~100%. Is that right? My intuition might just be mis-calibrated, but I would've thought the effect of the minor overlaps in sensitivity should be larger. (Maybe not though, one of the benefits of tetrachromacy...).

vis.flowers <- vismodel(flowers, visual = 'bluetit')
tcs.flowers <- colspace(vis.flowers)
summary(tcs.flowers)
plot(tcs.flowers, gamut = TRUE)

[Bisaloo]

All very good points:

1. Fixed by tryCatch() in 7a6be24
2. Updated the docs of triplot(), tetraplot() and tcsplot() in edd2c85
3. I think the code is correct. From a theoretical point of view (and actually one reviewer said the same thing), I agree but because of their oil droplets (this is the main reason I think), most birds have very little overlap. Below are the blue tit's photoreceptor sensitivity profiles:

The only case where it's impossible to excite only one receptor is for m (because there is always another receptor overlapping) and notice how it's the only place where the max gamut doesn't reach the corner in the tcs plot.

Let's try again with sensmodel(c(350, 450, 550, 650)), which generates a lot of overlap:

vis.flowers <- vismodel(flowers, sensmodel(c(350, 450, 550, 650)))
tcs.flowers <- colspace(vis.flowers)
summary(tcs.flowers)
plot(tcs.flowers, gamut = TRUE)

[thomased]

Cool - all looks good to go from my perspective then. Guess my intuition was just off re: birds, but it makes sense. Benefits of a finely tuned tetrachromat!