Replace values within a groupby based on multiple conditions

问题

My question is related to this one but I'm still not seeing how I can apply the answer to my problem. I have a DataFrame like so:

df = pd.DataFrame({
    'date': ['2001-01-01', '2001-02-01', '2001-03-01', '2001-04-01', '2001-02-01', '2001-03-01', '2001-04-01'],
    'cohort': ['2001-01-01', '2001-01-01', '2001-01-01', '2001-01-01', '2001-02-01', '2001-02-01', '2001-02-01'],
    'val': [100, 101, 102, 101, 200, 201, 201]
})

df
    date        cohort      val
0   2001-01-01  2001-01-01  100
1   2001-02-01  2001-01-01  101
2   2001-03-01  2001-01-01  102
3   2001-04-01  2001-01-01  101
4   2001-02-01  2001-02-01  200
5   2001-03-01  2001-02-01  201
6   2001-04-01  2001-02-01  201

Grouping for each cohort, I want to replace the values of val with the maximum value of val, but only for observations where date is less than the date associated with the maximum value of val. So rows 0, 1, and 4 would be changed to look like this:

df #This is what I want my final df to look like 
    date        cohort      val
0   2001-01-01  2001-01-01  102
1   2001-02-01  2001-01-01  102
2   2001-03-01  2001-01-01  102
3   2001-04-01  2001-01-01  101
4   2001-02-01  2001-02-01  201
5   2001-03-01  2001-02-01  201
6   2001-04-01  2001-02-01  201

How can I do this without lots of loops?

回答1:

Determine the maximum value of val PER GROUP of cohort
Determine the maximum date associated with val
Perform vectorised comparison and replacement with np.where

v = df.groupby('cohort').val.transform('max')
df['val'] = np.where(
    df.date <= df.set_index('cohort').val.idxmax(), v, df.val
)

df
    date        cohort      val
0   2001-01-01  2001-01-01  102
1   2001-02-01  2001-01-01  102
2   2001-03-01  2001-01-01  102
3   2001-04-01  2001-01-01  101
4   2001-02-01  2001-02-01  201
5   2001-03-01  2001-02-01  201
6   2001-04-01  2001-02-01  201

来源：https://stackoverflow.com/questions/50418372/replace-values-within-a-groupby-based-on-multiple-conditions

标签

python

pandas

dataframe