FacetGrid 上的 Seaborn 颜色条用于具有标准化颜色映射的 histplot

Question

我似乎无法使用 seaborn FacetGrid 显示二维 histplot 的颜色条。有人可以指出我缺少的链接吗？

了解到已经讨论过类似的解决方案，我无法适应我的用例：

1. Has the right position and values for color bar but isn't working for histplot
2. This proposal is not running at all & is rather dated so I am not sure it is still supposed to work
3. Seems to have fixed vmin/vmax and does not work with histplot

具体来说，我希望扩展下面的代码，以便显示颜色条。

import pandas as pd
import numpy as np
import seaborn as sns

df = pd.DataFrame(list(zip([random.randint(0,10) for i in range(1000)], pd.to_datetime(
                            [d.strftime('%Y-%m-%d') for d in pd.date_range('1800-01-01', periods=250, freq='1d')]+\
                            [d.strftime('%Y-%m-%d') for d in pd.date_range('1800-01-01', periods=250, freq='1d')]+\
                            [d.strftime('%Y-%m-%d') for d in pd.date_range('1800-01-01', periods=250, freq='1d')]+\
                            [d.strftime('%Y-%m-%d') for d in pd.date_range('1800-01-01', periods=250, freq='1d')]),
                            [random.choice(string.ascii_letters[26:30]) for i in range(1000)])), 
                            columns=["range","date","case_type"])
df["range"][df["case_type"]=="A"] = [random.randint(4562,873645) for i in range(1000)] 
df["range"][df["case_type"]=="C"] = [random.random() for i in range(1000)] 
fg = sns.FacetGrid(df, col="case_type", col_wrap=2, sharey=False)

fg.map(sns.histplot, "date", "range", stat="count", data=df)
fg.set_xticklabels(rotation=30)
fg.fig.show()

目标是在分面网格的右侧有一个颜色条，跨越整个图表 - 这里有两行，但可能会显示更多。显示的 2D 直方图具有一些非常不同的数据类型，因此每个 bin 和颜色的计数可能非常不同，知道“深蓝色”是 100 还是 1000 很重要。

编辑：为了清楚起见，从 cmets 看来，问题分为两个步骤：

1. 如何标准化所有绘图之间的颜色编码和
2. 使用标准化颜色映射在绘图右侧显示一个颜色条

Answer 1

我不确定是否有一种与生俱来的方式来实现您想要的情节。但是我们可以预先计算出bin number 和vmin/vmax 的合理值，并将它们应用于所有histplots：

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np

#generate a test dataset with different case_type probabilities
np.random.seed(123)
p1, p2, p3 = 0.8, 0.1, 0.03
df = pd.DataFrame(list(zip(np.random.randint(0, 20, 1000), 
                  pd.to_datetime(4 * [d.strftime('%Y-%m-%d') for d in pd.date_range('1800-01-01', periods=250, freq='1d')]),
                  np.random.choice(list("ABCD"),size=1000, p=[p1, p2, p3, 1-(p1+p2+p3)]))), 
                  columns=["range","date","case_type"])
df.loc[df.case_type == "A", "range"] *=   3
df.loc[df.case_type == "B", "range"] *=  23
df.loc[df.case_type == "C", "range"] *= 123

#determine the bin number for the x-axis
_, bin_edges = np.histogram(df["date"].dt.strftime("%Y%m%d").astype(int), bins="auto")
bin_nr = len(bin_edges)-1

#predetermine min and max count for each category
c_types = df["case_type"].unique()
vmin_list, vmax_list = [], []
for c_type in c_types:
    arr, _, _ = np.histogram2d(df.loc[df.case_type == c_type, "date"], df.loc[df.case_type == c_type, "range"], bins=bin_nr)
    vmin_list.append(arr.min())
    vmax_list.append(arr.max())
    
#find lowest and highest counts for all subplots
vmin_all = min(vmin_list)
vmax_all = max(vmax_list)

#now we are ready to plot
fg = sns.FacetGrid(df, col="case_type", col_wrap=2, sharey=False)
#create common colorbar axis
cax = fg.fig.add_axes([.92, .12, .02, .8])
#map colorbar to colorbar axis with common vmin/vmax values
fg.map(sns.histplot,"date", "range", stat="count", bins=bin_nr, vmin=vmin_all, vmax=vmax_all, cbar=True, cbar_ax=cax, data=df)
#prevent overlap
fg.fig.subplots_adjust(right=.9)
fg.set_xticklabels(rotation=30)

plt.show()

示例输出：

您可能还注意到我更改了您的示例数据框，以便 case_types 以不同的频率出现，否则您看不到 histplots 之间的太大差异。您还应该注意，histplots 是按照它们在数据框中出现的顺序绘制的，这可能不是您希望在图表中看到的顺序。

免责声明：这主要基于mwaskom's answer。