ENH: Accept no fields for groupby by

[simonaubertbd]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

Hello,

Sometimes, you have no fields to group by when aggregating. I know there is then an aggregate function but it would help having a more dynamic code to allow the use of groupy by without any grouping field instead of this error :

Best regards,

Simon

Feature Description

Just the ability to select no fields in the by argument

Alternative Solutions

A conditional function that uses groupby or aggregate

Additional Context

No response

[udit710]

take

[udit710]

Hi @simonaubertbd what's the expected output here? Assuming you have kept the alternative solution to use aggregate, do we need it to work the same as aggregate?

[simonaubertbd]

Hello @udit710 and thanks for your answer.

let's say I have

<style> </style>

bird	year	weight	age
Wilbur	1992	3	50
Donald Duck	1991	2	70
Scrooge McDuck	1993	2	100

aggregate1 = inlineInput1.groupby([]).agg(weight_max=('weight', 'max')).reset_index()

would be simply 3

[snitish]

Any thoughts on this @rhshadrach? I concur with OP that this feature would be helpful in cases where the grouping columns are dynamically determined.

[Delengowski]

So you want the groupby to be a no op and just return the dataframe if no grouping columns are specified? I sort of get it, you dont want to split the dataframe on anything so it just passes through.

I don't see how overloading the method and muddying the API here is worth while. Just do the check yourself. I think it would be strange in some cases for groupby to return a data frame and a group by object in another.

[snitish]

The resulting object can still be a GroupBy object, as @udit710 implemented in #61168.

[rhshadrach]

If I'm understanding the request right, @simonaubertbd desires for df.groupby([]).agg(...) to behave the same as df.agg(...). Here I am -1; the work it would take to get these to agree, and the ongoing maintenance to support, seems to me to be a non-starter.

On the other hand, I'm a bit more receptive to df.groupby([]) behaving the same as df.groupby(pd.Series(0, index=df.index)) (perhaps better would be to groupby np.zeros? Would have to run some benchmarks). However still here, it seems to me that having this live in user code rather than pandas is more explicit and readable.