2

I have some test data:

import numpy as np
x_data = np.arange(10)
y = np.random.rand(len(x_data))

With different properties

ix1 = x_data < 5
ix2 = x_data >= 5

I want to investigate the differences visually, but am messing the plot up:

import matplotlib.pyplot as plt
import seaborn as sns
sns.set_context('poster')

fig, ax = plt.subplots(figsize=(4, 4))
for i, x in enumerate(x_data):
    if ix1[i]:
        sns.set_palette('rainbow', sum(ix1))
    if ix2[i]:
        sns.set_palette('coolwarm', sum(ix2))
    plt.plot(x, y[i], 'o', label='{}'.format(x))
plt.legend(loc='best', prop={'size': 6})
plt.show()

The result should be points 0-4 are rainbow (red-violet) and points 5-9 are coolwarm (blue-white-red), but instead:

seaborn output

So, two questions:

  1. Is it ok to call sns.set_palette() after calling plt.subplots?
  2. Is there a way to set the palette more than once?
Serenity
  • 35,289
  • 20
  • 120
  • 115
jakebrinkmann
  • 704
  • 9
  • 23
  • I believe that setting the palette happens for an entire axis at once. That means it's not coloring each point as it's plotted, but instead, whichever palette is set last is applied to all the points. No idea how to make it do what you want though. – Wilduck Sep 17 '14 at 15:46

1 Answers1

2

No, because of the way matplotlib works, the color palette is a property of the Axes object and so whatever the currently set palette is at the time an Axes is created is what it's going to use. This is possible to get around if you want to hack on private attributes (see here), but I wouldn't really recommend that.

Here's what I could come up with in your case, using a somewhat different approach that might not be broadly applicable:

pal1 = sns.color_palette('rainbow', sum(ix1))
pal2 = sns.color_palette('coolwarm', sum(ix2))

fig, ax = plt.subplots(figsize=(4, 4))
ax.scatter(x_data[ix1], y[ix1], c=pal1, s=60, label="smaller")
ax.scatter(x_data[ix2], y[ix2], c=pal2, s=60, label="larger")
ax.legend(loc="lower right", scatterpoints=5)

enter image description here

FWIW, this visualization feels pretty complex and hard to process (and the two palettes you've chosen overlap a fair amount and aren't really appropriate for these data) so it might be worth starting with something simpler.

Community
  • 1
  • 1
mwaskom
  • 46,693
  • 16
  • 125
  • 127
  • Ahhh! I didn't even think to extract the palette and pass it to the plotting function. Maybe I'll take your suggestion about the overlap, and just create separate subplots to show the differences. Thank you Mark! – jakebrinkmann Sep 17 '14 at 16:37
  • @mwaskom Please kindly have a look at this problem https://stackoverflow.com/questions/44477755/how-to-control-the-color-of-graph-lines-in-matplotlib – Liza Jun 10 '17 at 23:15