如何为Plotly单个轨迹正确添加回调？三种方法问题排查

问题背景 #

我尝试用三种方式为Plotly轨迹注册点击回调：

• 方法'a'：将轨迹添加到图形前，直接在轨迹对象上调用on_click注册回调
• 方法'b'：将轨迹添加到图形后，从fig.data列表中获取对象并注册回调
• 方法'c'：方法'a'的变体，使用本地函数而非成员函数作为回调

实际运行后仅方法'b'生效，但存在缺陷：点击任意轨迹上的点时，所有轨迹的回调都会被触发，只能通过检查points.point_inds长度来规避无效触发。而按照官方文档实现的方法'a'和'c'完全不生效，请问操作哪里有误？

代码示例 #

import ipywidgets as widgets
import plotly.graph_objects as go

line = {'name': 'line','data': ((1,1), (2,2), (3,3)), 'color':'red', 'dash':'solid'}
squared = {'name': 'squared','data': ((1,1), (2,2**2), (3,3**2)), 'color':'blue', 'dash':'4,4'}
cubed = {'name': 'cubed','data': ((1,1), (2,2**3), (3,3**3)), 'color':'green', 'dash':'solid'}
n4 = {'name': 'n4','data': ((1,1), (2,2**4), (3,3**4)), 'color':'purple', 'dash':'solid'}
traces = (line, squared, cubed, n4)

class MyPlot:
    def __init__(self, traces, use_callback):
        self.traces = traces
        self.use_callback = use_callback
    
    def get_values(self, func, index):
        return [e[index] for e in func['data']]
    
    def callback_a(self, trace, points, state):
        print(f"in callback_a with trace = {trace}, points = {points}, state = {state}")
        
    def callback_b(self, trace, points, state):
        if len(points.point_inds) < 1:
            return
        print(f"in callback_b with trace = {trace}, points = {points}, state = {state}")
       
            
    def display(self):
        
        def callback_c(trace, points, state):
            print(f"in callback_c with trace = {trace}, points = {points}, state = {state}")
        
        fig = go.FigureWidget() 
               
        for t in traces:
            s = go.Scatter(mode="lines", name=t['name'], x=self.get_values(t, 0), y=self.get_values(t, 1), line=dict(width=2, color=t['color'], dash=t['dash']))
            if self.use_callback == 'a': s.on_click(self.callback_a)
            if self.use_callback == 'c': s.on_click(callback_c)
            fig.add_trace(s) 
            if self.use_callback == 'b': fig.data[-1].on_click(self.callback_b)
            fig.layout.title = f"Plot using callback {self.use_callback}"
            
        display(fig)
        
my_plot_a = MyPlot(traces, 'a')
my_plot_b = MyPlot(traces, 'b')
my_plot_c = MyPlot(traces, 'c')
my_plot_a.display()
my_plot_b.display()
my_plot_c.display()

问题原因分析 #

1. 方法'a'和'c'失效的核心原因：
   当你创建go.Scatter对象并调用on_click注册回调后，执行fig.add_trace(s)时，Plotly会创建该轨迹对象的深副本，原始对象上绑定的回调不会被复制到这个副本中。图形实际渲染和响应事件的是副本对象，所以原始对象上的回调根本不会被触发。

2. 方法'b'的回调批量触发问题：
   FigureWidget的点击事件处理机制是：先触发所有轨迹的on_click回调，再通过points.point_inds参数告知哪个轨迹实际被点击（只有被点击的轨迹对应的points.point_inds是非空列表）。这是Plotly的设计逻辑，并非代码错误，所以必须通过判断points.point_inds长度来过滤无效回调。

修正方案 #

方案1：统一用方法'b'的逻辑（确保回调注册到副本上） #

如果要保留单个轨迹绑定回调的方式，不管是成员函数还是本地函数，都要在轨迹被添加到图形后，针对fig.data里的副本对象注册回调，比如修正方法'a'和'c'的代码：

def display(self):
    def callback_c(trace, points, state):
        print(f"in callback_c with trace = {trace}, points = {points}, state = {state}")
    
    fig = go.FigureWidget() 
           
    for t in traces:
        s = go.Scatter(mode="lines", name=t['name'], x=self.get_values(t, 0), y=self.get_values(t, 1), line=dict(width=2, color=t['color'], dash=t['dash']))
        fig.add_trace(s) 
        if self.use_callback == 'a': fig.data[-1].on_click(self.callback_a)
        if self.use_callback == 'c': fig.data[-1].on_click(callback_c)
        if self.use_callback == 'b': fig.data[-1].on_click(self.callback_b)
        fig.layout.title = f"Plot using callback {self.use_callback}"
        
    display(fig)

方案2：使用全局点击回调（更简洁） #

直接给FigureWidget绑定全局点击回调，在回调内判断被点击的轨迹，避免逐个绑定：

def display(self):
    fig = go.FigureWidget() 
    for t in traces:
        s = go.Scatter(mode="lines", name=t['name'], x=self.get_values(t, 0), y=self.get_values(t, 1), line=dict(width=2, color=t['color'], dash=t['dash']))
        fig.add_trace(s) 

    def global_click_callback(trace, points, state):
        if len(points.point_inds) == 0:
            return
        print(f"点击轨迹：{trace.name}，点索引：{points.point_inds}")

    # 给所有轨迹绑定同一个全局回调
    for trace in fig.data:
        trace.on_click(global_click_callback)

    fig.layout.title = "全局回调演示"
    display(fig)

内容的提问来源于stack exchange，提问作者nPn