Symlog scale leads to weird results

[aplavin]

I’m trying to plot a 2d array with both positive an negative values, using a “symlog” colorscale in Makie.

This is what I get, looks weird – I cannot make any sense out of colorbar labels:

And this is what I expected it to be – result from matplotlib for the same array for comparison:

My issue here is with the colorscale specifically, not any other visual difference between Makie and matplotlib.

Full code for both:

# Makie:
let
	fig, ax, plt = image(A, colorscale=Makie.Symlog10(0.1), interpolate=false, colormap=:viridis)
	Colorbar(fig[1,2], plt)
	fig
end

# PyPlot:
let
	plt.figure()
	plt.imshow(A, norm=matplotlib.colors.SymLogNorm(0.1))
	plt.colorbar()
	plt.gcf()
end

• are you running newest version (version from docs) ? – yes
• can you reproduce the bug with a fresh environment ? (]activate --temp; add Makie) – yes
• What platform + GPU are you on? – Any backend, same result.

[jkrumbiegel]

ok so I think the labels are explained such that the linear tick finder is applied, which sees limits from about -2 to 1. So you get ticks -2, -1, 0, 1 but in the symlog scaling, -1, 0 and 1 for some reason all go on top of each other. That I don't understand

[aplavin]

Colors themselves also seem wrong in general...
Another random realization of the same code:
Makie

Matplotlib

[asinghvi17]

What happens if you try pseudolog?

[aplavin]

Pseudolog displays fine, but it's not symlog :) Most importantly, it lacks the parameter that decides on where log-scale effectively starts.
I think the issue is just in the Makie Symlog function, not sure where the formula is taken from – but appears very different from matlab/matplotlib at least.
Played with the symlog function I wrote quite some time ago for different purposes,

function symlog(x; linthresh, base=10., linscale=1.)
    linscale_adj = linscale / (1 - base^-1)
    if abs(x) < linthresh
        return x * linscale_adj
    else
        return sign(x) * linthresh * (linscale_adj + log(base, abs(x) / linthresh))
    end
end

– it seems to work fine as a scale in Makie.

[jkrumbiegel]

I don't remember anymore where I got the formula from, if I didn't come up with it myself from my understanding of what symlog should do. If it turns out to be incorrect, let's change it by all means.

[asinghvi17]

I believe the Symlog10 function actually constructs a Makie.ReversibleScale object - making that more extensible would be great as well! Maybe something like Symlog(linthresh, base = 10., linscale = 1.) which returns a ReversibleScale would be nice.