_shade_map normalizes vmin and vmax instead of recalculating them from normalized zdata

[maxhollmann]

Hey Raphael, thanks for this library, it's been very useful to me!

I noticed that the vmin & vmax values weren't respected when using shaders. The colorbar used the range set by set_plot_specs, but the colors drawn on the map itself were using the min/max of the normalized values. This fixes it by normalizing the existing vmin/vmax instead of recalculating them after normalization.

[raphaelquast]

Hey, @maxhollmann, thanks, nice to hear!
... and of course thanks a lot for taking the time to open a pull request!!

I had a quick look but I could not reproduce the behaviour you're describing...
Could you maybe add a reproducible example so that I can see what's going wrong?

Here's my quick test that does exactly what I would expect it to do:

code to reproduce

from eomaps import MapsGrid
import numpy as np

lon, lat = np.meshgrid(np.linspace(-89, 89, 500), np.linspace(-89, 89, 500))

mg = MapsGrid(1, 2, figsize=(10, 5), wspace=0, left=0.02)
mg.add_feature.preset.coastline()
mg.set_data(np.sqrt(lon**2 + lat**2), lon, lat)
mg.set_shape.shade_raster()
mg.set_plot_specs(vmin=25, vmax=100)

mg[0,0].set_classify_specs("EqualInterval", k=10)

mg.plot_map()
mg.add_colorbar(histbins=50)

for m in mg:
    m.cb.pick.attach.annotate()

[maxhollmann]

Here's an example that reproduces what I described:

from eomaps import Maps
import numpy as np

lon, lat = np.meshgrid(np.linspace(-89, 89, 500), np.linspace(-89, 89, 500))

data = np.stack([
    np.sqrt(lon**2 + lat**2) + i * 20
    for i in range(4)
])

vmin, vmax = np.min(data), np.max(data)

def prepare_layer(l, i):
    l.add_feature.preset.ocean()
    l.add_feature.preset.coastline()
    l.set_plot_specs(vmin=vmin, vmax=vmax)
    l.set_data(data[i, :, :], lon, lat)
    l.set_shape.shade_raster()
    l.plot_map()
    l.add_colorbar()

m = Maps(crs=4326)
prepare_layer(m, 0)
for i in range(1, data.shape[0]):
    layer = m.new_layer(layer=str(i))
    prepare_layer(layer, i)
    
m.util.layer_slider()

Not sure what makes the difference, I can look into it later today.

[raphaelquast]

ooh, i see 😅 yep you are totally right!
The current way for re-evaluating vmin/vmax from the normalized data only works as long as vmin & vmax are inside the data-range. In your case however they are outside the data-range (a perfectly valid use-case) and so using np.nanmin(normalized-data) does not respect the provided limits but always sets them to the lowest data-value.

Thanks for bringing this up!

[raphaelquast]

❗ Can you please rebase the branch to derive from the "dev" branch?

• I'm in the process of releasing v3.4 with a lot of updates and then I could include this right away
• also pre-commits require an update of the black version to run
  (which is why the tests would fail at the moment... this is already fixed in the dev branch)

[maxhollmann]

Done!

[raphaelquast]

wow that was fast! thanks!
sorry... but can you fetch the latest commit I did just right now?
(i realized that I had tests configured to trigger only on "push" and not on arbitrary "pull-requests" which is why they don't seem to run in here (you're the first external contributor so sorry for that 😅 )

[raphaelquast]

OK no need to do another rebase... a hotfix in the master-branch (and re-opening the pull-request) did the job equally well!

[codecov-commenter]

Codecov Report

Merging #64 (323e303) into dev (cbe1248) will not change coverage.
The diff coverage is 0.00%.

@@           Coverage Diff           @@
##              dev      #64   +/-   ##
=======================================
  Coverage   73.22%   73.22%           
=======================================
  Files          12       12           
  Lines        5323     5323           
=======================================
  Hits         3898     3898           
  Misses       1425     1425           

Impacted Files	Coverage Δ
eomaps/eomaps.py	70.97% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cbe1248...323e303. Read the comment docs.

[raphaelquast]

OK, all checks out fine. Thanks for your contribution!! It'll be in v3.4 (to be released today or tomorrow)

[maxhollmann]

Awesome, thanks :)