熊猫datetime系列的汇总仅作为datetime系列返回

Question

I have a dataframe like

test = pd.DataFrame({'date': ['2013-10-14 21:46:40', '2013-07-17 02:55:06', '2013-01-28 20:25:17'], 'category': [1, 1, 2]})
test['date'] = pd.to_datetime(test['date'])

     category                  date
0           1   2013-10-14 21:46:40
1           1   2013-07-17 02:55:06
2           2   2013-01-28 20:25:17

and I would like to compute some summary statistics for each category, specifically the earliest and latest date as well as the number of items in each category. The obvious way (to me) to do this is:

test.groupby('category')['date'].agg([len, min, max])

but when I do this, the len column gets automatically cast as np.datetime64 , which I assume is happening because that's the dtype of the original date column:

                                     len                   min                   max
category                                                                                    
1          1970-01-01 00:00:00.000000002   2013-07-17 02:55:06   2013-10-14 21:46:40
2          1970-01-01 00:00:00.000000001   2013-01-28 20:25:17   2013-01-28 20:25:17

I could go back and reconvert this len column to nanoseconds since GMT epoch, but that is pretty ugly and I feel like there must be a better way. Any ideas?

Answer 1

use 'size' ; this is currently an API bug (in that the len should just be translated directly to size ), see here

In [5]: test.groupby('category')['date'].agg(['size', min, max])
Out[5]: 
          size                 min                 max
category                                              
1            2 2013-07-17 02:55:06 2013-10-14 21:46:40
2            1 2013-01-28 20:25:17 2013-01-28 20:25:17