Seaborn：带 DivergingNorm/Cmap 的条形图

Question

我有一个范围从 -0.15 到 0.08 的值列表：

array([-9.28024375e-03, -7.74566792e-03,  6.89222284e-02,  1.98236910e-03,
        1.05891798e-02,  7.36737261e-03,  6.25777898e-03, -1.78726642e-02,
       -5.06597295e-03,  5.17623104e-02, -1.13474442e-02,  1.06263056e-02,
        2.09431952e-03, -1.54730073e-02, -1.93402164e-02,  1.04915526e-02,
        2.04725155e-03,  2.65222141e-02,  1.43185909e-02, -3.73984434e-03,
        2.62798866e-02, -2.67092615e-02,  3.48239927e-02,  3.08109938e-03,
       -9.12865632e-03,  2.46767319e-03, -2.36669926e-02,  2.07367834e-02,
        3.06733189e-02, -5.56772675e-03, -2.40482345e-03, -4.24432795e-02,
       -3.79769064e-03,  2.51791666e-02,  2.32164137e-02, -1.74955467e-02,
        7.47313626e-03,  6.86957861e-03,  1.38965986e-02,  7.68997312e-05,
       -4.59857112e-03,  1.37564169e-02, -6.25312715e-03,  1.66797351e-02,
       -7.13480355e-03, -2.38543967e-02,  2.48704615e-02,  2.99393285e-02,
       -1.17281194e-03,  1.78675678e-03,  8.04761250e-03, -1.50505912e-01,
        8.25650062e-02])

这些值中的每一个都对应于句子的特定部分。我试图在条形图中可视化这一点。我希望所有负值的范围从一些橙色 -> 红色，以及一些黄色 -> 绿色或蓝色 -> 深蓝色的正值。我希望这些值的阈值为 0.00。

我尝试使用 Matplotlib 的 DivergingNorm 功能来创建它并将其传递给我的 Seaborn 绘图。不幸的是，它似乎根本不起作用，我得到以下信息：

要创建它，我有以下代码：

import matplotlib.pyplot as plt
import matplotlib.colors as colors
import seaborn as sns

divnorm = colors.DivergingNorm(vmin=df["saliency"].values.min(), vcenter=0, vmax=df["saliency"].values.max())
div_colors = plt.cm.RdYlGn(divnorm(df["saliency"])) 
ax = sns.barplot(x='saliency', y='tokens', data=df, palette=div_colors, edgecolor='black')

我不确定我做错了什么，或者我是否应该使用不同的比例，但感谢任何帮助。

编辑：使用 DataFrame 值更新

            tokens  saliency
0               Go -0.009280
1               to -0.007746
2         Walmart,  0.068922
3              get  0.001982
4            shot.  0.010589
5               Go  0.007367
6               to  0.006258
7                a -0.017873
8           garlic -0.005066
9        festival,  0.051762
10             get -0.011347
11           shot.  0.010626
12              Go  0.002094
13              to -0.015473
14               a -0.019340
15        concert,  0.010492
16             get  0.002047
17           shot.  0.026522
18              Go  0.014319
19              to -0.003740
20         church,  0.026280
21             get -0.026709
22           shot.  0.034824
23              Go  0.003081
24              to -0.009129
25             the  0.002468
26         movies, -0.023667
27             get  0.020737
28           shot.  0.030673
29              Go -0.005568
30              to -0.002405
31           work, -0.042443
32             get -0.003798
33           shot.  0.025179
34              Go  0.023216
35              to -0.017496
36        college,  0.007473
37             get  0.006870
38           shot.  0.013897
39              Go  0.000077
40              to -0.004599
41         school,  0.013756
42             get -0.006253
43           shot.  0.016680
44              Go -0.007135
45              to -0.023854
46             the  0.024870
47            bar,  0.029939
48             get -0.001173
49           shot.  0.001787
50  #ThisIsAmerica  0.008048
51         #ElPaso -0.150506
52         #Dayton  0.082565

Answer 1

当一个标记有多个显着性时，seaborn 的条形图取这些显着性的平均值。因此，显示的不是原始显着性，而是每个标记的平均值以及误差线。

要为它们着色，您可以首先创建条形图，然后在第二次循环中遍历创建的条并根据它们的宽度为它们分配颜色。 （请注意，在最新的 matplotlib 版本中，DivergingNorm 已重命名为 TwoSlopeNorm。）

import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl
import seaborn as sns

saliency = np.array([-9.28024375e-03, -7.74566792e-03, 6.89222284e-02, 1.98236910e-03, 1.05891798e-02, 7.36737261e-03, 6.25777898e-03, -1.78726642e-02, -5.06597295e-03, 5.17623104e-02, -1.13474442e-02, 1.06263056e-02, 2.09431952e-03, -1.54730073e-02, -1.93402164e-02, 1.04915526e-02, 2.04725155e-03, 2.65222141e-02, 1.43185909e-02, -3.73984434e-03, 2.62798866e-02, -2.67092615e-02, 3.48239927e-02, 3.08109938e-03, -9.12865632e-03, 2.46767319e-03, -2.36669926e-02, 2.07367834e-02, 3.06733189e-02, -5.56772675e-03, -2.40482345e-03, -4.24432795e-02, -3.79769064e-03, 2.51791666e-02, 2.32164137e-02, -1.74955467e-02, 7.47313626e-03, 6.86957861e-03, 1.38965986e-02, 7.68997312e-05, -4.59857112e-03, 1.37564169e-02, -6.25312715e-03, 1.66797351e-02, -7.13480355e-03, -2.38543967e-02, 2.48704615e-02, 2.99393285e-02, -1.17281194e-03, 1.78675678e-03, 8.04761250e-03, -1.50505912e-01, 8.25650062e-02])
tokens = ['Go', 'to', 'Walmart,', 'get', 'shot.', 'Go', 'to', 'a', 'garlic', 'festival,', 'get', 'shot.', 'Go', 'to', 'a', 'concert,', 'get', 'shot.', 'Go', 'to', 'church,', 'get', 'shot.', 'Go', 'to', 'the', 'movies,', 'get', 'shot.', 'Go', 'to', 'work,', 'get', 'shot.', 'Go', 'to', 'college,', 'get', 'shot.', 'Go', 'to', 'school,', 'get', 'shot.', 'Go', 'to', 'the', 'bar,', 'get', 'shot.', '#ThisIsAmerica', '#ElPaso', '#Dayton']

ax = sns.barplot(x=saliency, y=tokens, edgecolor='black')

widths = np.array( [bar.get_width() for bar in ax.containers[0]])
divnorm = mpl.colors.TwoSlopeNorm(vmin=widths.min(), vcenter=0, vmax=widths.max())
div_colors = plt.cm.RdYlGn(divnorm(widths))
for bar, color in zip(ax.containers[0], div_colors):
    bar.set_facecolor(color)
plt.tight_layout()
plt.show()

现在，要从问题中获取颜色，您可以创建两个线性颜色图，一个用于负片，一个用于正片。

widths = np.array([bar.get_width() for bar in ax.containers[0]])
neg_cmap = mpl.colors.LinearSegmentedColormap.from_list('', ['orange', 'red'])
pos_cmap = mpl.colors.LinearSegmentedColormap.from_list('', ['yellow', 'green'])
min_width, max_width = widths.min(), widths.max()
for bar, w in zip(ax.containers[0], widths):
    bar.set_facecolor(neg_cmap(w / min_width) if w < 0 else pos_cmap(w / max_width))