如何在Seaborn FacetGrid分组柱状图中添加flag列自定义标注

实现思路 #

Seaborn的FacetGrid会按指定的row和col参数拆分数据生成子图，我们可以通过遍历每个子图对应的分组键，筛选原始DataFrame中对应的数据，再给每个柱子匹配flag列的内容并添加文本。

完整代码 #

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

df = pd.DataFrame([
    ['C', 'G1', 'gbt', 'auc', 0.7999, "†"],
    ['C', 'G1', 'gbtv2', 'auc', 0.8199, "*"],
    ['C', 'G1', 'gbt', 'pr@2%', 0.0883, "*"],
    ['C', 'G1', 'gbt', 'pr@10%', 0.0430, ""],
    ['C', 'G2', 'gbt', 'auc', 0.7554, ""],
    ['C', 'G2', 'gbt', 'pr@2%', 0.0842, ""],
    ['C', 'G2', 'gbt', 'pr@10%', 0.0572, ""],
    ['C', 'G3', 'gbt', 'auc', 0.7442, ""],
    ['C', 'G3', 'gbt', 'pr@2%', 0.0894, ""],
    ['C', 'G3', 'gbt', 'pr@10%', 0.0736, ""],
    ['E', 'G1', 'gbt', 'auc', 0.7988, ""],
    ['E', 'G1', 'gbt', 'pr@2%', 0.0810, ""],
    ['E', 'G1', 'gbt', 'pr@10%', 0.0354, ""],
    ['E', 'G1', 'gbtv3', 'pr@10%', 0.0454, ""],
    ['E', 'G2', 'gbt', 'auc', 0.7296, ""],
    ['E', 'G2', 'gbt', 'pr@2%', 0.1071, ""],
    ['E', 'G2', 'gbt', 'pr@10%', 0.0528, "†"],
    ['E', 'G3', 'gbt', 'auc', 0.6958, ""],
    ['E', 'G3', 'gbt', 'pr@2%', 0.1007, ""],
    ['E', 'G3', 'gbt', 'pr@10%', 0.0536, "†"],
  ], columns=["src", "grp", "model", "metric", "val", "flag"])

# 绘制FacetGrid柱状图
cat = sns.catplot(data=df, x="grp", y="val", hue="model", kind="bar", sharey=False, 
            col="metric", row="src")

# 遍历每个子图，添加flag文本
for row_idx, src_val in enumerate(cat.row_names):
    for col_idx, metric_val in enumerate(cat.col_names):
        # 获取当前子图的轴对象
        ax = cat.facet_axis(row_idx, col_idx)
        # 筛选当前子图对应的数据集
        subset = df[(df['src'] == src_val) & (df['metric'] == metric_val)]
        # 按grp和model排序，确保和柱子顺序一致
        subset = subset.sort_values(['grp', 'model'])
        # 遍历每个柱子和对应的flag
        for patch, flag in zip(ax.patches, subset['flag']):
            if flag:  # 只添加非空的flag
                # 计算柱子中心的x坐标，顶部的y坐标
                x = patch.get_x() + patch.get_width() / 2
                y = patch.get_height()
                # 添加文本
                ax.text(x, y, flag, ha='center', va='bottom', fontsize=12)

plt.show()

关键说明 #

• 通过cat.row_names和cat.col_names获取所有分组的键值，遍历每个子图对应的src和metric
• 筛选出当前子图对应的数据集后，必须按grp和model排序，保证和柱子的绘制顺序一致
• 利用patch.get_x()、patch.get_width()、patch.get_height()计算文本的位置，ha='center'确保水平居中，va='bottom'让文本在柱子顶部显示
• 只对非空的flag值添加文本，避免空字符串干扰

内容的提问来源于stack exchange，提问作者ironv