groupby.Transform Aggregation into Collections returns individual elements (unless added or string concatenated)

[yvanaquino]

import pandas as pd

dat = {
    'user': ['Aaron', 'Aaron', 'Bob', 'Bob', 'Cindy', 'Cindy'],
    'likes': ['Apples', 'Bananas', 'Cherries', 'Dates', 'eggs', 'fruit'],
    'vals': [1, 2, 3, 4, 5, 6]
}
df = pd.DataFrame(dat)

# This doesn't work and returns a like indexed NDFrame with each individual element.
df.groupby('user')['likes'].transform(list) 
df.groupby('user')['likes'].transform(lambda s: list(s))
df.groupby('user')['likes'].transform(pd.Series.tolist)
df.groupby('user')['likes'].transform(lambda s: pd.Series.tolist(s))


# This worked - but ideally, shouldn't require this much effort.  There seems to be a problem with Transform and iterables in the new API (I kind think i remember this working in .18...)

def pd_xform_list(_iter):
    tmpList = []
    for x in _iter:
        tmpList.append(x)
    return str(tmpList)
    
df['groups'] = df.groupby('user')['likes'].transform(lambda k: pd_xform_list(k))
from ast import literal_eval

df['groups'] = df['groups'].apply(lambda st: literal_eval(st))
df

When attempting to transform (pd.groupby.transform) into collections in pandas, the like NDFrame object returns the individual elements instead of a collection. string to literal evaluation workaround must be implemented. This functions normally with custom aggregation (groupby.agg) - but does not work correctly with groupby.transform.

Regards,

[gfyoung]

@yvanaquino : Thanks for reporting this! I'm a little unclear about your issue. Are you just requesting that we return a collection instead of individual elements when calling .transform? Because using .agg I think would easily solve your problem as you so noted above.

[yvanaquino]

Maybe I'm misunderstansing how to use DataFrame.transform() - shouldn't it return a broadcasted version as a likely indexed NDFrame of your aggregation?

In this example, I have to convert my list to a string and then apply literal evaluation from ast module to convert it back into an object.

Is there a way to handle the return a list with transform without numerical aggregation or string concatenation? IE - can it handle returning collections?

[javierquincke]

I think this is the same problem, https://stackoverflow.com/questions/62458837/groupby-transform-to-list-in-pandas-does-not-work