You need to enable JavaScript to run this app.
优惠活动
大模型
产品
解决方案
定价
更多

如何在含断裂X轴与双断裂Y轴的子图中绘制跨图LOESS趋势线

avatar
AI 小北
2026-6-12
带断裂轴的子图中绘制跨区域LOESS趋势线

问题背景

参考Matplotlib断裂Y轴的实现思路，基于合成数据搭建了包含断裂X轴与双断裂Y轴的子图布局：将多数常规数据放在下方主区域放大展示，异常值则置于上方的小区域。现需实现：使用sns.regplot绘制两条连续的局部加权（LOESS）趋势线，分别对应Cost与Hours字段，要求趋势线能跨越多张子图，适配Age>10的右侧区域以及上方异常值区域的维度。

合成数据 

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

np.random.seed(123)

# 生成合成数据：Age最小为1，Hours为正态分布，Cost为伽马分布缩放后的值
df = pd.DataFrame({
    "Age": np.random.poisson(lam=7, size=1000) + 1,
    "Hours": np.random.normal(loc=2100, scale=100, size=1000), 
    "Cost": np.random.gamma(shape=1.5, scale=1, size=1000) * 1000
})

当前实现代码（已完成布局与散点绘制） 

figure = plt.figure(figsize=(10, 5))

# 划分2行2列的网格，设置高度、宽度比例
full_grid = GridSpec(nrows=2, ncols=2, height_ratios=[1, 4], width_ratios=[4, 1])
ax1 = figure.add_subplot(full_grid[0, 0])  # 左上：Age<=10的异常值区域
ax2 = figure.add_subplot(full_grid[0, 1])  # 右上：Age>10的异常值区域
ax3 = figure.add_subplot(full_grid[1, 0])  # 左下：Age<=10的常规数据区域
ax4 = figure.add_subplot(full_grid[1, 1])  # 右下：Age>10的常规数据区域

# 轴关联规则：
# - ax1与ax3共享X轴范围；ax2与ax4共享X轴范围
# - ax3与ax4共享Y轴范围；ax1与ax2共享Y轴范围

# ---------------------- 下方主数据区域绘制 ----------------------
ax3hrs = ax3.twinx()  # 用于绘制Hours的副轴
ax4hrs = ax4.twinx()

# 绘制Age<=10的常规散点
sns.scatterplot(data=df[(df["Age"] <= 10) & (df["Cost"] <= df["Cost"].quantile(0.90))], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax3)
sns.scatterplot(data=df[(df["Age"] <= 10) & (df["Hours"] <= df["Hours"].quantile(0.90))], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax3hrs)

# 绘制Age>10的常规散点
sns.scatterplot(data=df[(df["Age"] > 10) & (df["Cost"] <= df["Cost"].quantile(0.90))], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax4)
sns.scatterplot(data=df[(df["Age"] > 10) & (df["Hours"] <= df["Hours"].quantile(0.90))], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax4hrs)

# 隐藏副轴的标签与刻度
ax3.set_ylabel(None); ax3.set_xlabel(None)
ax3hrs.set_ylabel(None); ax3hrs.set_yticklabels("")
ax3hrs.tick_params(axis="y", length=0)

# 隐藏多余边框
ax3.spines[["right", "top"]].set_visible(False)
ax3hrs.spines[["right", "top"]].set_visible(False)

ax4.set_ylabel(None); ax4.set_xlabel(None); ax4.set_yticklabels("")
ax4hrs.set_ylabel(None)
ax4.tick_params(axis="y", length=0)
ax4.spines[["left", "top"]].set_visible(False)
ax4hrs.spines[["left", "top"]].set_visible(False)

# 设置左下区域的X轴刻度
ax3.set_xticks(range(1, 11))
ax3.set_xticklabels(range(1, 11))

# 添加X轴断裂标记
d = 0.95  # 断裂斜线角度
break_kwargs = dict(marker=[(-1, -d), (1, d)], markersize=20,
                    linestyle="none", color='k', mec='k', mew=1, clip_on=False)
ax3.plot([1.015, 1.015], [0, 0.03], transform=ax3.transAxes, **break_kwargs)

# ---------------------- 上方异常值区域绘制 ----------------------
ax1hrs = ax1.twinx()
ax2hrs = ax2.twinx()

# 同步轴范围
ax1.set_xlim(ax3.get_xlim())
ax2.set_xlim(ax4.get_xlim())

# 绘制Age<=10的异常值散点
sns.scatterplot(data=df[(df["Age"] <= 10) & (df["Cost"] > df["Cost"].quantile(0.90))], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax1)
sns.scatterplot(data=df[(df["Age"] <= 10) & (df["Hours"] > df["Hours"].quantile(0.90))], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax1hrs)

# 绘制Age>10的异常值散点
sns.scatterplot(data=df[(df["Age"] > 10) & (df["Cost"] > df["Cost"].quantile(0.90))], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax2)
sns.scatterplot(data=df[(df["Age"] > 10) & (df["Hours"] > df["Hours"].quantile(0.90))], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax2hrs)

# 隐藏多余标签与边框
ax1.set_xlabel(None); ax1.set_xticklabels(""); ax1.tick_params(axis="x", length=0)
ax1.set_ylabel(None); ax1hrs.set_ylabel(None); ax1hrs.set_yticklabels("")
ax1hrs.tick_params(axis="y", length=0)
ax1.spines[["bottom", "right"]].set_visible(False)
ax1hrs.spines[["bottom", "right"]].set_visible(False)

ax2.set_xlabel(None); ax2.set_ylabel(None); ax2.set_xticklabels("")
ax2.set_yticklabels(""); ax2.tick_params(axis="both", length=0)
ax2hrs.set_xlabel(None); ax2hrs.set_ylabel(None)
ax2.spines[["left", "bottom"]].set_visible(False)
ax2hrs.spines[["left", "bottom"]].set_visible(False)

# 添加Y轴断裂标记
ax3.plot([0.015, 0.015], [1.095, 1.05], transform=ax3.transAxes, **break_kwargs)
ax4.plot([1.25, 1.25], [1.025, 1.075], transform=ax3.transAxes, **break_kwargs)

plt.subplots_adjust(wspace=0.025)

实现跨区域LOESS趋势线

核心思路

先对全量数据拟合LOESS模型，生成完整的趋势线坐标点；再根据各子图的轴范围，拆分趋势线数据，将对应片段绘制到对应的子图中，保证线条在断裂处视觉连续。

完整代码（含趋势线） 

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

np.random.seed(123)

# 生成合成数据
df = pd.DataFrame({
    "Age": np.random.poisson(lam=7, size=1000) + 1,
    "Hours": np.random.normal(loc=2100, scale=100, size=1000), 
    "Cost": np.random.gamma(shape=1.5, scale=1, size=1000) * 1000
})

# 预计算分位数，用于划分常规值与异常值
cost_q90 = df["Cost"].quantile(0.90)
hours_q90 = df["Hours"].quantile(0.90)

figure = plt.figure(figsize=(10, 5))
full_grid = GridSpec(nrows=2, ncols=2, height_ratios=[1, 4], width_ratios=[4, 1])
ax1 = figure.add_subplot(full_grid[0, 0])
ax2 = figure.add_subplot(full_grid[0, 1])
ax3 = figure.add_subplot(full_grid[1, 0])
ax4 = figure.add_subplot(full_grid[1, 1])

# ---------------------- 拟合LOESS趋势线，生成全量趋势数据 ----------------------
# 拟合Cost的LOESS模型，获取趋势线数据
cost_reg = sns.regplot(data=df, x="Age", y="Cost", lowess=True, scatter=False, ax=plt.gca())
cost_line = cost_reg.get_lines()[0]
x_cost, y_cost = cost_line.get_data()
plt.close()  # 关闭临时绘图

# 拟合Hours的LOESS模型，获取趋势线数据
hours_reg = sns.regplot(data=df, x="Age", y="Hours", lowess=True, scatter=False, ax=plt.gca())
hours_line = hours_reg.get_lines()[0]
x_hours, y_hours = hours_line.get_data()
plt.close()

# ---------------------- 下方主数据区域 ----------------------
ax3hrs = ax3.twinx()
ax4hrs = ax4.twinx()

# 绘制散点
sns.scatterplot(data=df[(df["Age"] <=10) & (df["Cost"] <= cost_q90)], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax3)
sns.scatterplot(data=df[(df["Age"] <=10) & (df["Hours"] <= hours_q90)], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax3hrs)
sns.scatterplot(data=df[(df["Age"] >10) & (df["Cost"] <= cost_q90)], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax4)
sns.scatterplot(data=df[(df["Age"] >10) & (df["Hours"] <= hours_q90)], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax4hrs)

# 绘制Cost趋势线的常规值片段
mask_cost_regular = (x_cost <=10) & (y_cost <= cost_q90)
ax3.plot(x_cost[mask_cost_regular], y_cost[mask_cost_regular], color="darkorange", linewidth=2)
mask_cost_regular_right = (x_cost >10) & (y_cost <= cost_q90)
ax4.plot(x_cost[mask_cost_regular_right], y_cost[mask_cost_regular_right], color="darkorange", linewidth=2)

# 绘制Hours趋势线的常规值片段
mask_hours_regular = (x_hours <=10) & (y_hours <= hours_q90)
ax3hrs.plot(x_hours[mask_hours_regular], y_hours[mask_hours_regular], color="darkblue", linewidth=2)
mask_hours_regular_right = (x_hours >10) & (y_hours <= hours_q90)
ax4hrs.plot(x_hours[mask_hours_regular_right], y_hours[mask_hours_regular_right], color="darkblue", linewidth=2)

# 轴样式调整
ax3.set_ylabel(None); ax3.set_xlabel(None)
ax3hrs.set_ylabel(None); ax3hrs.set_yticklabels(""); ax3hrs.tick_params(axis="y", length=0)
ax3.spines[["right", "top"]].set_visible(False); ax3hrs.spines[["right", "top"]].set_visible(False)
ax4.set_ylabel(None); ax4.set_xlabel(None); ax4.set_yticklabels("")
ax4hrs.set_ylabel(None); ax4.tick_params(axis="y", length=0)
ax4.spines[["left", "top"]].set_visible(False); ax4hrs.spines[["left", "top"]].set_visible(False)
ax3.set_xticks(range(1,11)); ax3.set_xticklabels(range(1,11))

# 添加X轴断裂标记
d = 0.95
break_kwargs = dict(marker=[(-1, -d), (1, d)], markersize=20,
                    linestyle="none", color='k', mec='k', mew=1, clip_on=False)
ax3.plot([1.015,1.015], [0,0.03], transform=ax3.transAxes, **break_kwargs)

# ---------------------- 上方异常值区域 ----------------------
ax1hrs = ax1.twinx()
ax2hrs = ax2.twinx()
ax1.set_xlim(ax3.get_xlim()); ax2.set_xlim(ax4.get_xlim())

# 绘制散点
sns.scatterplot(data=df[(df["Age"] <=10) & (df["Cost"] > cost_q90)], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax1)
sns.scatterplot(data=df[(df["Age"] <=10) & (df["Hours"] > hours_q90)], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax1hrs)
sns.scatterplot(data=df[(df["Age"] >10) & (df["Cost"] > cost_q90)], 
                x="Age", y="Cost", color="grey", alpha=0.1, ax=ax2)
sns.scatterplot(data=df[(df["Age"] >10) & (df["Hours"] > hours_q90)], 
                x="Age", y="Hours", color="grey", alpha=0.1, ax=ax2hrs)

# 绘制Cost趋势线的异常值片段
mask_cost_outlier = (x_cost <=10) & (y_cost > cost_q90)
ax1.plot(x_cost[mask_cost_outlier], y_cost[mask_cost_outlier], color="darkorange", linewidth=2)
mask_cost_outlier_right = (x_cost >10) & (y_cost > cost_q90)
ax2.plot(x_cost[mask_cost_outlier_right], y_cost[mask_cost_outlier_right], color="darkorange", linewidth=2)

# 绘制Hours趋势线的异常值片段
mask_hours_outlier = (x_hours <=10) & (y_hours > hours_q90)
ax1hrs.plot(x_hours[mask_hours_outlier], y_hours[mask_hours_outlier], color="darkblue", linewidth=2)
mask_hours_outlier_right = (x_hours >10) & (y_hours > hours_q90)
ax2hrs.plot(x_hours[mask_hours_outlier_right], y_hours[mask_hours_outlier_right], color="darkblue", linewidth=2)

# 轴样式调整
ax1.set_xlabel(None); ax1.set_xticklabels(""); ax1.tick_params(axis="x", length=0)
ax1.set_ylabel(None); ax1hrs.set_ylabel(None); ax1hrs.set_yticklabels(""); ax1hrs.tick_params(axis="y", length=0)
ax1.spines[["bottom", "right"]].set_visible(False); ax1hrs.spines[["bottom", "right"]].set_visible(False)
ax2.set_xlabel(None); ax2.set_ylabel(None); ax2.set_xticklabels("")
ax2.set_yticklabels(""); ax2.tick_params(axis="both", length=0)
ax2hrs.set_xlabel(None); ax2hrs.set_ylabel(None)
ax2.spines[["left", "bottom"]].set_visible(False); ax2hrs.spines[["left", "bottom"]].set_visible(False)

# 添加Y轴断裂标记
ax3.plot([0.015,0.015], [1.
相关产品推荐
方舟 Agent Plan

超全模态模型 × Harness 升级，最新支持 Deepseek-V4.1-Flash、GLM-5.3 系列、Doubao-Seedream-5.0-pro、Kimi-K3 (部分)， 限时 9.9 元起

产品介绍相关文档
相关产品推荐
方舟 Agent Plan

超全模态模型 × Harness 升级，最新支持 Deepseek-V4.1-Flash、GLM-5.3 系列、Doubao-Seedream-5.0-pro、Kimi-K3 (部分)， 限时 9.9 元起

产品介绍相关文档
火山引擎 最新活动
全面开放
Doubao-Seedance-2.5全新模型升级

模型再升级,30秒超长叙事, 模态参考扩容

HOT
方舟Codingplan

模型自由，工具不限，最新支持 Deepseek-V4 系列、GLM-5.3 系列

HOT
方舟Agent plan

超全模态模型 × Harness 升级，最新支持 Deepseek-V4.1-Flash、GLM-5.3 系列

HOT
新用户特惠专场

大模型19元起，Al应用9.9元畅享，新人首购爆款尽享优惠

最近更新时间：2026.06.12 12:01:38