为什么pandas中的`mean`在处理Series时有效，但在处理GroupBy对象时无效？

huangapple
go评论174阅读模式
英文:

Why does pandas mean, on datetime, work on a series but not on a groupby object

问题

我正在尝试按组计算日期的均值。

  1. import pandas as pd
  3. df = pd.DataFrame({'Id': ['A', 'A', 'B', 'B'],
  4.                    'Date': [pd.datetime(2000, 12, 31), pd.datetime(2002, 12, 31),
  5.                             pd.datetime(2000, 6, 30), pd.datetime(2002, 6, 30)]})

这一直是一个让人头疼的问题，所以我很高兴地了解到这似乎在pandas 0.25中已经修复了 https://stackoverflow.com/questions/27907902/datetime-objects-with-pandas-mean-function。

  1. df['Date'].mean()
  2. Out[45]: Timestamp('2001-09-30 00:00:00') # 这个可以工作

然而，使用groupby无法做到这一点。

  1. df.groupby('Id')['Date'].mean()
  3. Traceback (most recent call last):
  5.   File "<ipython-input-46-5fae5ffac6c6>", line 1, in <module>
  6.     df.groupby('Id')['Date'].mean()
  8.   File "C:\Users\xxx\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\groupby\groupby.py", line 1205, in mean
  9.     "mean", alt=lambda x, axis: Series(x).mean(**kwargs), **kwargs
  11.   File "C:\Users\xxx\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\groupby\groupby.py", line 888, in _cython_agg_general
  12.     raise DataError("No numeric types to aggregate")
  14. DataError: No numeric types to aggregate
  16. 发生了什么情况，是否有一个简单的解决方法？
  18. <details>
  19. <summary>英文:</summary>
  21. I am trying to take the mean of dates in by groups.
  23.     import pandas as pd
  25.     df = pd.DataFrame({&#39;Id&#39;: [&#39;A&#39;, &#39;A&#39;, &#39;B&#39;, &#39;B&#39;],
  26.                        &#39;Date&#39;: [pd.datetime(2000, 12, 31), pd.datetime(2002, 12, 31),
  27.                                 pd.datetime(2000, 6, 30), pd.datetime(2002, 6, 30)]})
  29. This has always been a pain to do, so I was pleased to learn that this had apparntly been fixed in pandas 0.25 https://stackoverflow.com/questions/27907902/datetime-objects-with-pandas-mean-function.
  31.     df[&#39;Date&#39;].mean()
  32.     Out[45]: Timestamp(&#39;2001-09-30 00:00:00&#39;) # This works
  34. However, this cant be done using &#180;groupby&#180;
  36.     df.groupby(&#39;Id&#39;)[&#39;Date&#39;].mean()
  38.     Traceback (most recent call last):
  40.       File &quot;&lt;ipython-input-46-5fae5ffac6c6&gt;&quot;, line 1, in &lt;module&gt;
  41.         df.groupby(&#39;Id&#39;)[&#39;Date&#39;].mean()
  43.       File &quot;C:\Users\xxx\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\groupby\groupby.py&quot;, line 1205, in mean
  44.     &quot;mean&quot;, alt=lambda x, axis: Series(x).mean(**kwargs), **kwargs
  46.       File &quot;C:\Users\xxx\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\groupby\groupby.py&quot;, line 888, in _cython_agg_general
  47.     raise DataError(&quot;No numeric types to aggregate&quot;)
  49.     DataError: No numeric types to aggregate
  51. What is going on here, and is there an easy workaround?
  53. </details>
  56. # 答案1
  57. **得分**: 2
  59. 使用lambda函数与[`GroupBy.agg`](http://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.core.groupby.GroupBy.agg.html)或[`GroupBy.apply`](http://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.core.groupby.GroupBy.apply.html)：
  61. ```python
  62. print(df.groupby('Id')['Date'].agg(lambda x: x.mean()))
  63. print(df.groupby('Id')['Date'].agg(pd.Series.mean))
  64. print(df.groupby('Id')['Date'].apply(lambda x: x.mean()))
  65. print(df.groupby('Id')['Date'].apply(pd.Series.mean))
  67. Id
  68. A   2001-12-31
  69. B   2001-06-30
  70. Name: Date, dtype: datetime64[ns]

区别在于如果有多个列：

  1. df = pd.DataFrame({'Id': ['A', 'A', 'B', 'B'],
  2.                    'Date': [pd.datetime(2000, 12, 31), pd.datetime(2002, 12, 31),
  3.                             pd.datetime(2000, 6, 30), pd.datetime(2002, 6, 30)]})
  4. df['Date1'] = df['Date']
  5. print(df.groupby('Id').agg(lambda x: x.mean()))
  6.          Date      Date1
  7. Id                      
  8. A  2001-12-31 2001-12-31
  9. B  2001-06-30 2001-06-30
  10. print(df.groupby('Id').agg(pd.Series.mean))
  11.          Date      Date1
  12. Id                      
  13. A  2001-12-31 2001-12-31
  14. B  2001-06-30 2001-06-30
  16. print(df.groupby('Id').apply(lambda x: x.mean()))
  17. Empty DataFrame
  18. Columns: []
  19. Index: []
  21. print(df.groupby('Id').apply(pd.Series.mean))
  22. Empty DataFrame
  23. Columns: []
  24. Index: []

为什么pandas中的datetime上的mean在Series上有效，但在groupby对象上无效？

一段时间以前，对于Series、Datetime的mean存在问题，可以查看这里，所以在pandas的一些未来版本中，这个问题可能已经解决。

英文:

Use lambda function with GroupBy.agg or GroupBy.apply:

  1. print (df.groupby(&#39;Id&#39;)[&#39;Date&#39;].agg(lambda x: x.mean()))
  2. print (df.groupby(&#39;Id&#39;)[&#39;Date&#39;].agg(pd.Series.mean))
  3. print (df.groupby(&#39;Id&#39;)[&#39;Date&#39;].apply(lambda x: x.mean()))
  4. print (df.groupby(&#39;Id&#39;)[&#39;Date&#39;].apply(pd.Series.mean))
  5. Id
  6. A   2001-12-31
  7. B   2001-06-30
  8. Name: Date, dtype: datetime64[ns]

Difference is if multiple columns:

  1. df = pd.DataFrame({&#39;Id&#39;: [&#39;A&#39;, &#39;A&#39;, &#39;B&#39;, &#39;B&#39;],
  2. &#39;Date&#39;: [pd.datetime(2000, 12, 31), pd.datetime(2002, 12, 31),
  3. pd.datetime(2000, 6, 30), pd.datetime(2002, 6, 30)]})
  4. df[&#39;Date1&#39;] = df[&#39;Date&#39;]
  5. print (df.groupby(&#39;Id&#39;).agg(lambda x: x.mean()))
  6. Date      Date1
  7. Id                      
  8. A  2001-12-31 2001-12-31
  9. B  2001-06-30 2001-06-30
  10. print (df.groupby(&#39;Id&#39;).agg(pd.Series.mean))
  11. Date      Date1
  12. Id                      
  13. A  2001-12-31 2001-12-31
  14. B  2001-06-30 2001-06-30
  15. print (df.groupby(&#39;Id&#39;).apply(lambda x: x.mean()))
  16. Empty DataFrame
  17. Columns: []
  18. Index: []
  19. print (df.groupby(&#39;Id&#39;).apply(pd.Series.mean))
  20. Empty DataFrame
  21. Columns: []
  22. Index: []

>Why does pandas mean, on datetime, work on a series but not on a groupby object

Some time ago it was problem with mean for Series, Datetimes, check this, so possible in some next versions of pandas this should working well.

huangapple
  • 本文由 发表于 2020年1月6日 19:05:37
  • 转载请务必保留本文链接：https://go.coder-hub.com/59610966.html
  • datetime
  • mean
  • pandas
  • python
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定