GLMakie vs Plots, not enogh precision for axis limits

General Usage
, ,
1

I have ugly data with an ugly code, but I just used scatter in both places.

Plots gives this:
Screenshot from 2024-03-06 17-26-08

And, as you can see, the oscillations are in the double order.

GLMakie gives this:

Screenshot from 2024-03-06 17-24-54

If you zoom manually all the data behaves in the same way, however I can’t manually set it each time. Is there a way to say to GLMakie to not give up in setting axis limits?

P.S: I included a simple code:

using GLMakie

x = [1, 2]
y = Float64[1.00000001, 1.00000002]
f = Figure()
ax = Axis(f[1, 1])
scatter!(x, y)
display(f)

And GLMakie just gives up:
Screenshot from 2024-03-06 17-50-28

Found this:

2

You could try tightlimits!(ax) to force makie to zoom in around the points.

Regarding your example, I think the issue is that Makie converts everything to Float32 and so you lose the precision

Float32(1.00000001)
1.0f0

You can manually get around it by doing something like this:

x = [1, 2]
y = [1.00000001, 1.00000002]
yscaled = (y .- 1) .* 10^8

f = Figure()
ax = Axis(f[1, 1])
scatter!(x, yscaled)
ax.ytickformat=values->["$(round(value/10^8 + 1,digits=8))" for value in values]
display(f)

which gives this:

image

3

I can’t do that in my case. It was just a simple example. I need to rewrite everything in Plots, as nothing in Makie supports Float64 as ticks

4

My suggestion would be to make some data transformation before plotting. For example, subtract the mean value and then divide by the new mean value, to get a proper scaling for the comparison.

5

While I get that it’s annoying not to have the full Float64 range available right now for plotting in Makie, I also think that often plots that go to the tiny or huge float values are pretty hard to read. So for visualizations that are not one-offs where you quickly want to see some data, the scaling step is usually bearable. There is some work ongoing in Makie to improve the situation, but I don’t know when it will come to fruition.

6

Oh no, they are quite easy to read.

I found a solution here with the pythonplot package: