使用NetworkX时如何获取FancyArrowPatch端点的精确位置

问题 #

我用NetworkX和matplotlib复现指定图形，已创建好网络结构：

graph = nx.Graph()
elist = [(0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 0), (5, 1), (5, 2)]
graph.add_edges_from(elist)
pos = {0: (-1, 2), 1: (2, 2), 2: (3, 0.5), 3: (2, -1.5), 4: (-1.5, -1), 5: (0, 0)}

我想先绘制节点再单独绘制边，从边的位置获取端点来做散点标记，但用FancyArrowPatch绘制带箭头的边时，调用FancyArrowPatch.get_path()拿不到端点的精确位置。相关代码如下：

fig, ax = plt.subplots(figsize=(8, 6), dpi=200)
ax.set_axis_off()

node_object = nx.draw_networkx_nodes(graph, pos=pos, node_size=3000, node_color="#d3d3d3")
edge_object = nx.draw_networkx_edges(
    graph,
    pos=pos,
    node_size=10000,
    edge_color="#808080",
    width=2,
    arrows=True
)

node_object.set_zorder(-1)
for el in edge_object:
    paths = el.get_path().vertices[0], el.get_path().vertices[2]
    x, y = np.transpose(paths)
    middle = (paths[0] + paths[1]) / 2
    dist = np.abs(paths[1] - paths[0])
    
    print(middle, dist)
    print(el, x, y, fig.dpi)


    ax.scatter(x, y, s=100, c="#3a649d", edgecolors="#808080", linewidths=2)

请问问题原因是什么？有没有办法获取端点的精确位置？

问题原因 #

1. get_path()返回局部坐标路径：FancyArrowPatch的get_path()方法返回的是箭头补丁自身局部坐标系下的路径，并非画布或数据坐标系的全局位置。这些顶点坐标是相对补丁自身原点的，无法直接对应实际的边端点。
2. 边端点被自动偏移：NetworkX的draw_networkx_edges在绘制带箭头的边时，会根据node_size参数自动调整边的端点位置，让边停在节点边缘而非完全覆盖节点。此时FancyArrowPatch的实际端点是偏移后的位置，但get_path()无法直接输出这个全局坐标。

解决方法 #

方法1：手动计算偏移后端点 #

利用节点大小和原始坐标，计算边端点的偏移位置：

import networkx as nx
import matplotlib.pyplot as plt
import numpy as np

graph = nx.Graph()
elist = [(0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 0), (5, 1), (5, 2)]
graph.add_edges_from(elist)
pos = {0: (-1, 2), 1: (2, 2), 2: (3, 0.5), 3: (2, -1.5), 4: (-1.5, -1), 5: (0, 0)}

fig, ax = plt.subplots(figsize=(8, 6), dpi=200)
ax.set_axis_off()

node_size = 3000
node_object = nx.draw_networkx_nodes(graph, pos=pos, node_size=node_size, node_color="#d3d3d3")
edge_object = nx.draw_networkx_edges(
    graph,
    pos=pos,
    node_size=node_size,
    edge_color="#808080",
    width=2,
    arrows=True
)

node_object.set_zorder(-1)

for u, v in graph.edges():
    # 获取原始节点坐标
    x1, y1 = pos[u]
    x2, y2 = pos[v]
    # 计算节点半径（node_size为面积，半径需转换为数据坐标系单位）
    node_radius = np.sqrt(node_size / np.pi) / fig.dpi * ax.get_window_extent().width
    # 计算边的向量与长度
    dx = x2 - x1
    dy = y2 - y1
    length = np.sqrt(dx**2 + dy**2)
    # 计算偏移量，让边端点停在节点边缘
    if length > 0:
        offset_x = dx * node_radius / length
        offset_y = dy * node_radius / length
        start = (x1 + offset_x, y1 + offset_y)
        end = (x2 - offset_x, y2 - offset_y)
        # 绘制端点散点
        ax.scatter([start[0], end[0]], [start[1], end[1]], s=100, c="#3a649d", edgecolors="#808080", linewidths=2)

plt.show()

方法2：通过变换矩阵转换局部坐标 #

利用FancyArrowPatch的变换矩阵，将局部路径坐标转换为数据坐标系的全局坐标：

import networkx as nx
import matplotlib.pyplot as plt
import numpy as np

graph = nx.Graph()
elist = [(0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 0), (5, 1), (5, 2)]
graph.add_edges_from(elist)
pos = {0: (-1, 2), 1: (2, 2), 2: (3, 0.5), 3: (2, -1.5), 4: (-1.5, -1), 5: (0, 0)}

fig, ax = plt.subplots(figsize=(8, 6), dpi=200)
ax.set_axis_off()

node_object = nx.draw_networkx_nodes(graph, pos=pos, node_size=3000, node_color="#d3d3d3")
edge_object = nx.draw_networkx_edges(
    graph,
    pos=pos,
    node_size=10000,
    edge_color="#808080",
    width=2,
    arrows=True
)

node_object.set_zorder(-1)

for el in edge_object:
    # 获取箭头局部路径与变换矩阵
    path = el.get_path()
    transform = el.get_transform()
    # 将局部顶点转换为数据坐标系坐标
    data_vertices = transform.transform(path.vertices)
    # 箭头起点为第一个顶点，终点为倒数第二个顶点（最后一个为箭头尖端）
    start = data_vertices[0]
    end = data_vertices[-2]
    # 绘制端点散点
    ax.scatter([start[0], end[0]], [start[1], end[1]], s=100, c="#3a649d", edgecolors="#808080", linewidths=2)

plt.show()

内容的提问来源于stack exchange，提问作者Alessandro Romancino