使用 .groupby 的结果在 Pandas 中过滤和选择数据框

Question

Hello Python community,

I have a "little" Python/Pandas problem and would be very happy if someone could help me with it on short notice. I have a dataframe with 2 IDs, date, hour of day and several metrics like in this example:

index	ID_1	ID_2	date	hour	metric_1	metric_2	metric_3	…	metric_k
0	A101321	25459379	2021-05-09	0	11	any value	any value	…	any value
1	A101321	25459379	2021-05-09	1	8	any value	any value	…	any value
2	A101321	25459379	2021-05-09	2	7	any value	any value	…	any value
3	A101321	25459379	2021-05-09	3	0	any value	any value	…	any value
…	A101321	25459379	2021-05-09	0	0	any value	any value	…	any value
22	A101321	25459379	2021-05-09	22	17	any value	any value	…	any value
23	A101321	25459379	2021-05-09	23	11	any value	any value	…	any value
24	A101321	25459379	2021-05-10	0	9	any value	any value	…	any value
25	A101321	25459379	2021-05-10	1	3	any value	any value	…	any value
…	…	…	…	…	…	…	…	…	…
n	K510325	105983-20	2021-05-23	0	5	any value	any value	…	any value
n+1	K510325	105983-20	2021-05-23	1	1	any value	any value	…	any value

For each metric, one value is determined per device per day per hour. A device is made unique with 2 IDs because one ID is not unique. Now I want to know per device and per day at which hour eg metric_1 reaches the maximum value to see a distribution of hours with maximum values.

Using df[['ID_1', 'ID_2', 'date', 'metric_1']].groupby(['ID_1', 'ID_2', 'date'], as_index=False).max() I do get the maximum value of the day for a device displayed:

	ID_1	ID_2	date	metric_1
index
0	A101321	25459379	2021-05-09	17
1	A101321	25459379	2021-05-10	9
...	...	...	...	...
m	K510325	105983-20	2021-05-23	5

but I can't see at what hour and all attempts to achieve this have failed miserably so far... Can someone please help me with this?

Answer 1

If you take your groupby, which seems to produce the results you want, instead use

df[['ID_1', 'ID_2', 'date', 'metric_1']].groupby(['ID_1', 'ID_2', 'date']).idxmax()

This gets the indices for those maximum values. Then subset in your original data frame for those specific indices using df.loc[indicies, list_of_columns] , if the result of the above code block is assigned to indicies .