Select rows from dataframe based on a unique values of other column?

Question

One of the column of my dataframe has values shown as below:

air_voice_no_null.loc[:,"host_has_profile_pic"].value_counts(normalize = True)*100

1.0    99.694276
0.0     0.305724
Name: host_has_profile_pic, dtype: float64

that's 99:1 for each unique value in that column.

I now want to make a new a dataframe, such that it has 60% of 1.0 and 40% of 0.0 from that dataframe along with all rows(of course with less number of rows).

I've tried splitting it using strat function of train_test_split class of sklearn.model_selection as shown below, but with no luck getting dataframes with equal proportions of each unique value.

from sklearn.model_selection import train_test_split

profile_train_x, profile_test_x, profile_train_y, profile_test_y = train_test_split(air_voice_no_null.loc[:,['log_price', 'accommodates', 'bathrooms','host_response_rate', 'number_of_reviews', 'review_scores_rating','bedrooms', 'beds', 'cleaning_fee', 'instant_bookable']],
                                                                                   air_voice_no_null.loc[:,"host_has_profile_pic"],
                                                                                   random_state=42, stratify=air_voice_no_null.loc[:,"host_has_profile_pic"])

and this is what the above code resulted in, with no change in number of rows.

print(profile_train_x.shape)
print(profile_test_x.shape)
print(profile_train_y.shape)
print(profile_test_y.shape)

(55442, 10)
(18481, 10)
(55442,)
(18481,)

How do I select subset of my dataset with a decreased number of rows, while maintaining appropriate proportions of each class of the host_has_profile_pic variable.

link to the complete dataset: https://www.kaggle.com/stevezhenghp/airbnb-price-prediction

Answer 1

Consider the following way:

import pandas as pd

# create some data
df = pd.DataFrame({'a': [0] * 10 + [1] * 90})

print('original proportion:')
print(df['a'].value_counts(normalize=True))

# take samples for every unique value separately
df_new = pd.concat([
    df[df['a'] == 0].sample(frac=.4),
    df[df['a'] == 1].sample(frac=.07)])

print('\nsample proportion:')
print(df_new['a'].value_counts(normalize=True))

Output:

original proportion:
1    0.9
0    0.1
Name: a, dtype: float64

sample proportion:
1    0.6
0    0.4
Name: a, dtype: float64