I have the following figure:

The figure is composed by the following code snippet:

fig = plt.figure(constrained_layout=True)
grid = fig.add_gridspec(2, 2)

ax_samples_losses = fig.add_subplot(grid[0, 0:])
ax_samples_losses.set_title('Avg. loss per train sample (epoch 0 excluded)')
for sample_idx, sample_avg_train_loss_history in enumerate(samples_avg_train_loss_history):
    ax_samples_losses.plot(sample_avg_train_loss_history, label='Sample '   str(sample_idx))
ax_samples_losses.set_title('Avg. loss per train sample (epoch 0 excluded)')
ax_samples_losses.set_xlabel('Epoch')
ax_samples_losses.set_ylabel('Sample avg. loss')
ax_samples_losses.set_xticks(range(1, epochs))
ax_samples_losses.tick_params(axis='x', rotation=90)
ax_samples_losses.yaxis.set_ticks(np.arange(0, np.max(samples_avg_train_loss_history), 0.25))
ax_samples_losses.tick_params(axis='both', which='major', labelsize=6)
plt.legend(bbox_to_anchor=(1, 1), prop={'size': 6}) #loc="upper left"
# fig.legend(...)

ax_patches_per_sample = fig.add_subplot(grid[1, 0])
#for sample_idx, sample_patches_count in enumerate(samples_train_patches_count):
#    ax_patches_per_sample.bar(sample_patches_count, label='Sample '   str(sample_idx))
ax_patches_per_sample.bar(range(0, len(samples_train_patches_count)), samples_train_patches_count, align='center')
ax_patches_per_sample.set_title('Patches per sample')
ax_patches_per_sample.set_xlabel('Sample')
ax_patches_per_sample.set_ylabel('Patch count')
ax_patches_per_sample.set_xticks(range(0, len(samples_train_patches_count)))
ax_patches_per_sample.yaxis.set_ticks(np.arange(0, np.max(samples_train_patches_count), 20))
ax_patches_per_sample.tick_params(axis='both', which='major', labelsize=6)

where

• samples_train_patches_count is a simple list with the number of patches per sampled image
• samples_avg_train_loss_history is a list of lists in the shape samples, epochs (so if viewed as a matrix every row will be a sample and every column will be the loss of that sample over time)

I do believe I need to do both

• shared legend
• shared color palette

The shared legend can be done by using get_legend_handles_labels(). However I do not know how to share colors. Both subplots describe different properties of the same thing - the samples. In short I would like to have Patches per sample subplot have all the colors Avg. loss per train sample (epoch 0 excluded) uses.

CodePudding user response：

The first plot is using standard matplotlib Tab10 discrete color map. We can create a cycler over this colormap, and set one by one the color of each bar:

import matplotlib.pyplot as plt
import matplotlib.cm as cm
from matplotlib.gridspec import GridSpec
import numpy as np
from itertools import cycle

# create a cycler to continously loop over a discrete colormap
cycler = cycle(cm.tab10.colors)

N = 10
x = np.arange(N).astype(int)
y = np.random.uniform(5, 15, N)

f = plt.figure()
gs = GridSpec(2, 4)
ax0 = f.add_subplot(gs[0, :-1])
ax1 = f.add_subplot(gs[1, :-1])
ax2 = f.add_subplot(gs[:, -1])

for i in x:
    ax0.plot(x, np.exp(-x / (i   1)), label="Sample %s" % (i   1))
h, l = ax0.get_legend_handles_labels()

ax1.bar(x, y)
for p in ax1.patches:
    p.set_facecolor(next(cycler))

ax2.axis(False)
ax2.legend(h, l)
plt.tight_layout()

EDIT to accommodate comment. To avoid repetitions you should use a colormap. Matplotlib offers many