Seaborn FacetGrid叠加箱线图与散点图：分类错位及可变X轴问题

问题原因 #

使用FacetGrid.map_dataframe绘制boxplot时，seaborn会基于整个数据集的Treatment变量全局类别集合设置所有子图的X轴刻度位置。当某个子图缺失特定类别数据（比如示例中的Patient2-Cohort2-Treatment2），boxplot仍会为全局所有类别预留位置，但实际绘制时仅用现有数据填充对应位置，导致后续类别（如Treatment3）的箱线图被错放到缺失类别的标签下。而stripplot是基于每个子图的实际数据动态生成点的位置，因此不会出现错位。

解决方案 #

方法1：自定义绘图函数，基于子图本地数据设置boxplot顺序 #

通过自定义函数，在每个子图绘制时提取当前数据中实际存在的Treatment类别，将其作为boxplot的order参数，确保箱线图与X轴标签匹配，同时只显示现有类别：

#!/usr/bin/env python3
import seaborn as sns
import pandas as pd
import numpy as np

# 数据生成逻辑与原代码一致
patients   = ['Patient1', 'Patient2', 'Patient3']
cohorts    = ['Cohort1', 'Cohort2', 'Cohort3']
treatments = ['Treatment1', 'Treatment2', 'Treatment3']

data = {
    'Patient': [],
    'Cohort': [],
    'Treatment': [],
    'Value': []
}

for patient in patients:
    for cohort in cohorts:
        for treatment in treatments:
            for i in range(10):
                data['Patient'].append(patient)
                data['Cohort'].append(cohort)
                data['Treatment'].append(treatment)
                data['Value'].append(np.random.rand())

df = pd.DataFrame(data)

index_to_drop = df[(df['Patient'] == 'Patient2') &
                   (df['Cohort'] == 'Cohort2') &
                   (df['Treatment'] == 'Treatment2')].index
df = df.drop(index_to_drop)

# 自定义子图绘制函数
def plot_boxstrip(data, x, y):
    # 筛选当前子图实际存在的Treatment类别，保留原始顺序
    current_treatments = [t for t in treatments if t in data[x].unique()]
    sns.boxplot(data=data, x=x, y=y, showfliers=False, order=current_treatments)
    sns.stripplot(data=data, x=x, y=y, jitter=True, order=current_treatments, color='black')

# 创建FacetGrid并应用自定义函数
facet_params = dict(
    data=df,
    col='Patient',
    row='Cohort',
    col_order=patients,
    row_order=cohorts
)
grid = sns.FacetGrid(**facet_params)
grid.map_dataframe(plot_boxstrip, x='Treatment', y='Value')

grid.tight_layout()
grid.savefig('facet_fixed.png')

方法2：改用seaborn.catplot（更简洁） #

catplot是seaborn的高级接口，内部会自动处理每个子图的分类变量级别，无需额外配置即可避免错位，同时默认只显示当前子图存在的类别：

#!/usr/bin/env python3
import seaborn as sns
import pandas as pd
import numpy as np

# 数据生成逻辑与原代码一致
patients   = ['Patient1', 'Patient2', 'Patient3']
cohorts    = ['Cohort1', 'Cohort2', 'Cohort3']
treatments = ['Treatment1', 'Treatment2', 'Treatment3']

data = {
    'Patient': [],
    'Cohort': [],
    'Treatment': [],
    'Value': []
}

for patient in patients:
    for cohort in cohorts:
        for treatment in treatments:
            for i in range(10):
                data['Patient'].append(patient)
                data['Cohort'].append(cohort)
                data['Treatment'].append(treatment)
                data['Value'].append(np.random.rand())

df = pd.DataFrame(data)

index_to_drop = df[(df['Patient'] == 'Patient2') &
                   (df['Cohort'] == 'Cohort2') &
                   (df['Treatment'] == 'Treatment2')].index
df = df.drop(index_to_drop)

# 使用catplot绘制箱线图，再叠加散点图
g = sns.catplot(
    data=df,
    col='Patient',
    row='Cohort',
    col_order=patients,
    row_order=cohorts,
    kind='box',
    x='Treatment',
    y='Value',
    showfliers=False
)
g.map_dataframe(sns.stripplot, x='Treatment', y='Value', jitter=True, color='black')

g.tight_layout()
g.savefig('catplot_fixed.png')

效果说明 #

两种方案均可解决箱线图数据错位问题，且每个子图的X轴仅显示当前子图存在的Treatment类别，无空白位置。其中catplot方法代码更简洁，推荐优先使用。

内容的提问来源于stack exchange，提问作者xApple