Moving average in pandas with condition-CodePudding

I have a dataframe with the following structure:

import numpy as np
import pandas as pd

df = pd.DataFrame(
    {
        "date": ["2020-01-01", "2020-01-02", "2020-01-03", "2020-01-04"] * 2,
        "group": ["A", "A", "A", "A", "B", "B", "B", "B"],
        "x": [1, 2, 2, 3, 2, 3, 4, 2],
        "condition": [1, 0, 1, 0] * 2
    }
)
df

I want to calculate, the rolling moving average of the last 3 days of the column x:

Per group
Using only past data (not using the current row)
Using only data for the rolling average where condition = 1.

The outcome should be the following:

How can I do that in pandas? Thanks!

Keep in mind it is not the same as this:

Rolling function in pandas with condition

In here I'm looking for the moving average of the last 3 days, in the other one I just wanted the rolling average.

CodePudding user response：

First replace not matched rows by NaN by Series.where and then per groups shift values and call rolling method:

f = lambda x: x.shift().rolling(3, min_periods=1).mean()
df['roll'] = (df.assign(x = df['x'].where(df['condition'].eq(1)))
                .groupby('group')['x']
                .transform(f))
print (df)
         date group  x  condition  roll
0  2020-01-01     A  1          1   NaN
1  2020-01-02     A  2          0   1.0
2  2020-01-03     A  2          1   1.0
3  2020-01-04     A  3          0   1.5
4  2020-01-01     B  2          1   NaN
5  2020-01-02     B  3          0   2.0
6  2020-01-03     B  4          1   2.0
7  2020-01-04     B  2          0   3.0

Details:

print (df.assign(x = df['x'].where(df['condition'].eq(1))))
         date group    x  condition
0  2020-01-01     A  1.0          1
1  2020-01-02     A  NaN          0
2  2020-01-03     A  2.0          1
3  2020-01-04     A  NaN          0
4  2020-01-01     B  2.0          1
5  2020-01-02     B  NaN          0
6  2020-01-03     B  4.0          1
7  2020-01-04     B  NaN          0