ENH: No numeric_only argument for pandas.core.groupby.GroupBy.rank()

[notBillJames]

pandas.core.groupby.GroupBy.rank does not have a numeric_only argument like DataFrame.rank()

I have a DataFrame with several statistics from baseball teams across different years. I want to rank each team in each statistic, grouped by season. Every column is numeric, except for the teamID column which is an object type containing the names of each team as a string. My code looks something like this

stats = pd.read_csv('stats.csv')
ranked = stats.groupby('yearID').rank(method='average', ascending=False)
ranked['teamID'] = stats['teamID']

Since it is GroupBy.rank() I can't pass the numeric_only argument and that means I have to reassign ranked['teamID'] to the original column. I also cannot do

ranked = stats.groupby(['yearID', 'teamID']).rank(...)

because that would give everybody a rank of 1.

Is there a reason that numeric_only is included in DataFrame.rank() but not GroupBy.rank(). Could it be added to GroupBy.rank()?

Then I could code it like this, which would be easier.

```python
stats = pd.read_csv('stats.csv')
ranked = stats.groupby('yearID').rank(method='average', ascending=False, numeric_only=False)

I am just a hobbyist and so I don't know much about the implementation of these methods which means there may be something I am completely ignoring, or another more efficient way to do it. If so I would appreciate some enlightenment about what I am missing. Thanks!

[Mislav-Ro]

I think I understood what you want but to be sure could you provide sample data set and what you are trying to accomplish? Perhaps provide 2 tables (in and out)

You are missing 1 step. You should specify what to do with the grouped values, should it take max value out of each groupby or min value etc.

pd.DataFrame.groupby().max().rank() --- I think this is what you are looking for

code below for your case:

ranked = stats.groupby('yearID').max().rank(method='average', ascending=False)

Hope this clarifies things a bit

P.S. you can also specify
pd.DataFrame.groupby().max().rank(numeric_only=True)

[notBillJames]

Basically I want to know how a team ranked for each statistic in a given season.
Going in the table would look like this:

	yearID	teamID	W	IP	K%	K/BB+	K%+	WAR
0	1947	Athletics	15.5	251.65	9.45	91.5	97	2.9
1	1947	Braves	21	277.6	11.1	155.5	116.5	5.5
2	1947	Browns	8	185.1	11.3	143	116	3.2
3	1947	Cardinals	15	196.05	12.5	168.5	131.5	4.45
4	1947	Cubs	10.5	195	10.25	119	107.5	2.55

Then going out I would like the table to look like this:

	yearID	teamID	W	IP	K%	K/BB+	K%+	WAR
0	1947	Athletics	2	2	5	5	5	4
1	1947	Braves	1	1	3	2	2	1
2	1947	Browns	5	5	2	3	3	3
3	1947	Cardinals	3	3	1	1	1	2
4	1947	Cubs	4	4	4	4	4	5

So the 1947 Braves finished first in W, IP, and WAR. My plan is to make a new column which is the average of all the other rankings and then sort by the average to see who was the most "dominant" in those areas. Would doing
ranked = stats.groupby('yearID').max().rank(method='average', ascending=False) accomplish that?

I did find a workaround by doing

ranked = stats.set_index(['yearID', 'teamID'])
ranked = ranked.groupby('yearID').rank(ascending=False).reset_index()

and that yields the same as above.

[Mislav-Ro]

I see.
Indexes do get erased when calling groupby().rank() , also numeric_only cannot be specified in that manner. We could try to implement this.

Your workaround looks fine.

Could I get assigned for this issue after someone else also confirms it?

[notBillJames]

Okay, thank you for the help!