不同采样率Timedelta索引Pandas DataFrame绘图报错排查

我通过以下代码将多个带Timedelta索引的Pandas DataFrame绘制到共享X轴上，这些DataFrame的采样率各不相同：

# 数据源
list_of_dfs = [instance.parent_perspective.associated[expression] for expression in instance.parent_perspective.associated]

figure, axis = plt.subplots(len(list_of_dfs), 1, sharex=True)

for dataframe in list_of_dfs:
    print(dataframe)

for i, dataframe in enumerate(list_of_dfs):
    ax = dataframe.plot(ax=axis[i])

当按采样率从低到高排序DataFrame列表时，代码能正常运行；但按采样率从高到低排序时，会触发如下错误：

File "C:\Users\[me]\..\[my_script].py", line 110, in [my_script]
    ax = dataframe.plot(ax=axis[i])
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_core.py", line 1030, in __call__
    return plot_backend.plot(data, kind=kind, **kwargs)
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_matplotlib\__init__.py", line 71, in plot
    plot_obj.generate()
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_matplotlib\core.py", line 501, in generate
    self._make_plot(fig)
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_matplotlib\core.py", line 1544, in _make_plot
    newlines = plotf(
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_matplotlib\core.py", line 1589, in _ts_plot
    freq, data = maybe_resample(data, ax, kwds)
  File "C:\Users\[me]\AppData\Roaming\Python\Python310\site-packages\pandas\plotting\_matplotlib\timeseries.py", line 84, in maybe_resample
    series.index = series.index.asfreq(  # type: ignore[attr-defined]
AttributeError: 'TimedeltaIndex' object has no attribute 'asfreq'. Did you mean: 'freq'?

数据示例 #

采样率从低到高排序（正常运行）

value
0 days 00:00:00           0.0
0 days 00:00:00.010000    0.0
0 days 00:00:00.020000    0.0
0 days 00:00:00.030000    0.0
0 days 00:00:00.040000    0.0
...                       ...
0 days 00:10:24.950000    0.0
0 days 00:10:24.960000    0.0
0 days 00:10:24.970000    0.0
0 days 00:10:24.980000    0.0
0 days 00:10:24.990000    0.0

[62500 rows x 1 columns]
                        value
0 days 00:00:00           0.0
0 days 00:00:00.010000    0.0
0 days 00:00:00.020000    0.0
0 days 00:00:00.030000    0.0
0 days 00:00:00.040000    0.0
...                       ...
0 days 00:10:24.950000    0.1
0 days 00:10:24.960000    0.1
0 days 00:10:24.970000    0.1
0 days 00:10:24.980000    0.1
0 days 00:10:24.990000    0.1

[62500 rows x 1 columns]
                             value
-1 days +23:59:59.999100  0.031557
-1 days +23:59:59.999200  0.032379
-1 days +23:59:59.999300  0.031419
-1 days +23:59:59.999400  0.032287
-1 days +23:59:59.999500  0.031857
...                            ...
0 days 00:10:24.998600    6.252964
0 days 00:10:24.998700    6.251458
0 days 00:10:24.998800    6.251105
0 days 00:10:24.998900    6.248893
0 days 00:10:24.999000    6.248886

[6250000 rows x 1 columns]

采样率从高到低排序（触发报错）

value
-1 days +23:59:59.999100  0.031557
-1 days +23:59:59.999200  0.032379
-1 days +23:59:59.999300  0.031419
-1 days +23:59:59.999400  0.032287
-1 days +23:59:59.999500  0.031857
...                            ...
0 days 00:10:24.998600    6.252964
0 days 00:10:24.998700    6.251458
0 days 00:10:24.998800    6.251105
0 days 00:10:24.998900    6.248893
0 days 00:10:24.999000    6.248886

[6250000 rows x 1 columns]
                        value
0 days 00:00:00           0.0
0 days 00:00:00.010000    0.0
0 days 00:00:00.020000    0.0
0 days 00:00:00.030000    0.0
0 days 00:00:00.040000    0.0
...                       ...
0 days 00:10:24.950000    0.1
0 days 00:10:24.960000    0.1
0 days 00:10:24.970000    0.1
0 days 00:10:24.980000    0.1
0 days 00:10:24.990000    0.1

[62500 rows x 1 columns]
                        value
0 days 00:00:00           0.0
0 days 00:00:00.010000    0.0
0 days 00:00:00.020000    0.0
0 days 00:00:00.030000    0.0
0 days 00:00:00.040000    0.0
...                       ...
0 days 00:10:24.950000    0.0
0 days 00:10:24.960000    0.0
0 days 00:10:24.970000    0.0
0 days 00:10:24.980000    0.0
0 days 00:10:24.990000    0.0

问题分析 #

报错根源在于Pandas的matplotlib绘图模块在处理共享X轴时，会尝试对后续绘制的DataFrame进行自动重采样以匹配已绘制轴的频率，但TimedeltaIndex并不支持asfreq()方法（该方法仅适用于DatetimeIndex）。当先绘制高采样率DataFrame后，低采样率DataFrame触发了这个无效的重采样逻辑，导致报错。

解决方案 #

1. 关闭自动重采样
   在调用plot()时添加resample=False参数，禁止Pandas自动重采样：

   for i, dataframe in enumerate(list_of_dfs):
       ax = dataframe.plot(ax=axis[i], resample=False)
   

2. 转换索引类型
   将TimedeltaIndex转换为DatetimeIndex，这样就能支持asfreq()方法。可以选择一个基准时间（比如pd.Timestamp('2020-01-01')），将Timedelta与基准时间相加得到Datetime：

   base_time = pd.Timestamp('2020-01-01')
   for df in list_of_dfs:
       df.index = base_time + df.index
   

3. 保持低采样率优先排序
   如果不需要改变排序顺序，继续使用采样率从低到高的排序方式，即可避免触发该错误。

内容的提问来源于stack exchange，提问作者Alrie