Seaborn箱线图异常：添加类别后原有箱线图被改写求助

我尝试基于不同气旋类型（V0_V1、V1、V0_V2_12、V2_12）绘制Na浓度的箱线图，但加入V0_V2_12、V2_12类别后，原有V0_V1、V1的箱线图发生变化，疑似被改写。代码如下：

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

# Load Excel file
file_path = 'XRF Co-locations.xlsx' # file name 
sheet = "Masterlist Co-locations" # sheet I want to isolate
df_dict = pd.read_excel(file_path, sheet_name=[sheet], header=2) # put into dictionary 
df = df_dict[sheet] 

# Clean column names
df.columns = df.columns.str.strip() # get rid of white spaces 

# Filter based on Notes
v0_v1_na = df[df["Notes"] == "V0_V1"]["Na (ng/cm2)"]
v1_na = df[df["Notes_2"] == "V1"]["Na (ng/cm2)_2"]

v0_v2_12_na = df[df["Notes"] == "V0_V2_12"]["Na (ng/cm2)"]
v2_12_na = df[df["Notes_2"] == "V2_12"]["Na (ng/cm2)_2"]

# Combine into DataFrames and compute median for each
df_mean_v0 = v0_na.median()  # calculate the median of v0_na
df_mean_v1 = v1_na.median()  # calculate the median of v1_na

df_mean_v0_v2_12 = v0_v2_12_na.median()  # calculate the median of Na for v0_v2_12
df_mean_v2_12 = v2_12_na.median()  # calculate the median of Na for v2_12

print(df_mean_v0)
print(df_mean_v1)
    
print(df_mean_v0_v2_12)
print(df_mean_v2_12)

# Plot 
plt.figure(figsize=(12, 6))

# Create boxplots for v0_na and v1_na side by side
sns.boxplot(data=[v0_na, v1_na, v0_v2_12_na, v2_12_na], orient="v", palette="husl")

# Adding labels, title, and customization
plt.ylabel("Na Concentration (ng/cm²)", fontsize=12)
plt.xlabel("Cyclone Type", fontsize=12)
plt.xticks([0, 1, 2, 3], ['V0_V1', 'V1', 'V0_V2_12', 'V2_12'], rotation=0)  # Custom x-ticks for the two categories
plt.title("Box Plot of Na (ng/cm²) for V0 and V1 Cyclones", fontsize=14)

plt.legend()
plt.show()

问题分析与修复步骤 #

1. 修正变量名错误 #

代码存在未定义变量问题：你定义了v0_v1_na，但后续计算中位数和绘图时误用了v0_na，这会直接报错，也可能干扰你对数据是否被改写的判断。先把所有v0_na替换为v0_v1_na。

2. 重构为长格式DataFrame #

当前用列表传入多个Series给sns.boxplot的方式，容易因不同类别数据范围差异导致坐标轴自动缩放，让原有箱线图视觉上“变形”（实际数据未修改，仅坐标轴范围变化）。改用长格式DataFrame能更清晰管理数据，避免这类问题：

# 重构为长格式数据
data_list = []
# 添加各类型数据（过滤空值）
data_list.extend([("V0_V1", val) for val in v0_v1_na.dropna()])
data_list.extend([("V1", val) for val in v1_na.dropna()])
data_list.extend([("V0_V2_12", val) for val in v0_v2_12_na.dropna()])
data_list.extend([("V2_12", val) for val in v2_12_na.dropna()])

# 转为标准DataFrame
plot_df = pd.DataFrame(data_list, columns=["Cyclone Type", "Na Concentration (ng/cm²)"])

# 绘制箱线图
plt.figure(figsize=(12, 6))
sns.boxplot(data=plot_df, x="Cyclone Type", y="Na Concentration (ng/cm²)", palette="husl")
plt.title("Box Plot of Na (ng/cm²) by Cyclone Type", fontsize=14)
plt.show()

3. 验证数据未被改写 #

先单独打印V0_V1和V1的统计量，确认添加新类别前后数据无变化：

# 打印V0_V1统计信息
print("V0_V1 统计量：")
print(v0_v1_na.describe())
# 打印V1统计信息
print("\nV1 统计量：")
print(v1_na.describe())

如果中位数、四分位数等统计量与之前一致，说明数据未被改写，只是坐标轴缩放导致视觉变化。

4. 手动控制坐标轴范围（可选） #

若新类别数据范围远大于原有类别，可手动设置y轴范围，保持原有箱线图的视觉比例：

plt.ylim(0, 自定义最大值)

内容的提问来源于stack exchange，提问作者Kathryn Cheng