Filter Pandas MultiIndex over all First Levels Columns

Question

Trying to find a way of efficiently filtering all entries under both top level columns based on a filter defined for only one of the top level columns. Best explained with the example below and desired output.

Example DataFrame

import pandas as pd
import numpy as np
info = ['price', 'year']
months = ['month0','month1','month2']
settlement_dates = ['2020-12-31', '2021-01-01']
Data = [[[2,4,5],[2020,2021,2022]],[[1,4,2],[2021,2022,2023]]]
Data = np.array(Data).reshape(len(settlement_date),len(months) * len(info))
midx = pd.MultiIndex.from_product([assets, Asset_feature])
df = pd.DataFrame(Data, index=settlement_dates, columns=midx)
df

            price                 year              
           month0 month1 month2 month0 month1 month2
2020-12-31      2      4      5   2020   2021   2022
2021-01-01      1      4      2   2021   2022   2023

Create filter for multiindex dataframe

idx_cols = pd.IndexSlice

df_filter = df.loc[:, idx_cols['year', :]]==2021

df[df_filter]


            price                  year               
           month0 month1 month2  month0  month1 month2
2020-12-31    NaN    NaN    NaN     NaN  2021.0    NaN
2021-01-01    NaN    NaN    NaN  2021.0     NaN    NaN

Desired output:

            price                  year               
           month0 month1 month2  month0  month1 month2
2020-12-31    NaN    4      NaN     NaN  2021.0    NaN
2021-01-01    1      NaN    NaN  2021.0     NaN    NaN

Answer 1

You can reshape for simplify solution by reshape for DataFrame by DataFrame.stack with filter by DataFrame.where:

df1 = df.stack()

df_filter = df1['year']==2021

df_filter = df1.where(df_filter).unstack()
print (df_filter)
            price                  year               
           month0 month1 month2  month0  month1 month2
2020-12-31    NaN    4.0    NaN     NaN  2021.0    NaN
2021-01-01    1.0    NaN    NaN  2021.0     NaN    NaN

Your solution is possible, but more complicated - there is reshaped mask for repalce missing values by back and forward filling missing values:

idx_cols = pd.IndexSlice

df_filter = df.loc[:, idx_cols['year', :]]==2021

df_filter = df_filter.reindex(df.columns, axis=1).stack(dropna=False).bfill(axis=1).ffill(axis=1).unstack()
print (df_filter)
            price                 year              
           month0 month1 month2 month0 month1 month2
2020-12-31  False   True  False  False   True  False
2021-01-01   True  False  False   True  False  False

print (df[df_filter])
            price                  year               
           month0 month1 month2  month0  month1 month2
2020-12-31    NaN    4.0    NaN     NaN  2021.0    NaN
2021-01-01    1.0    NaN    NaN  2021.0     NaN    NaN