Pandas aggregating and applying a condition on column names

I've got this DataFrame

And what I want to achieve is getting this dict which language as ak ey and the name of the is_columns a,d aggregating the ids

 {
  "de" :{"is_bruiser" : [121,1542], "is_tank" : [121,542 },
  "fr": {"is_bruiser" : [1216,4578] } 
  }

Have been trying few solutions but my bottleneck so far is applying the pre aggregation condition on the column names

You can apply a function to each group. Dictionary with result can be calculated for each group. And them grouped dataframe can be formatted as a final dictionary.

from collections import defaultdict

def get_dict(group):
    dct = defaultdict(list)
    bool_cols = filter(lambda x: x.startswith("is"), group.columns)
    
    for column in bool_cols:
        for _, row in group.iterrows():
            if row[column]:
                dct[column].append(row["id"])
    return dict(dct)
        
aggregated_df = df \
    .groupby("language") \
    .apply(get_dict)

dict(zip(aggregated_df.index, aggregated_df))

Output:

{'de': {'is_bruiser': ['121', '1542'], 'is_tank': ['121', '542']},
 'fr': {'is_bruiser': ['4578', '1216']}}

Another solution:

d1 = dict()
for language in df['language'].unique():
    df_filtered = df[df['language'] == language]
    d2 = dict()
    for col in ['is_bruiser', 'is_tank', 'is_support']:
        values = list(df_filtered[df_filtered[col]]['id'])
        if values:
            d2[col] = values
    d1[language] = d2

you can iterate in your languages and your boolean columns :

result = {} for lang in df["language"].unique():
    result[lang] = {}
    for col in ["is_bruiser", "is_tank", "is_support"]:
        ids = df.loc[(df["language"] == lang) & (df[col]), "id"].tolist()
        if ids:
            result[lang][col] = ids

Then you'll get what you want :

print(result)
> {'fr': {'is_bruiser': [4578, 1216]}, 'de': {'is_bruiser': [121, 1542], 'is_tank': [121, 542]}}