ENH: Enable rolling.apply on custom function that requires multiple columns of data frame

[bbkaran]

Is your feature request related to a problem?

Yes.

I wish I could use pandas to:

create an additional column 'newc' of my dataframe df as df['newc'] through rolling.apply on df['cond'] with a custom function. The custom function requires two columns of df.

Describe the solution you'd like

I prefer the solution to look something like:

df['newc'] = df['cond'].rolling(4).apply(T_correction, args = (df['temp'].rolling(4)))

where

>>> df.head()
                       temp   cond
ts
2018-06-01 00:00:00  51.908  27.83
2018-06-01 00:05:00  52.144  27.83
2018-06-01 00:10:00  51.880  27.83
2018-06-01 00:15:00  52.001  27.83
2018-06-01 00:20:00  51.835  27.83

def T_correction(df, d):
    df = pd.DataFrame(data = df)
    df.columns = ['cond']
    df['temp'] = d
    X = df.drop(['cond'], axis = 1)    # X features: temp

    X = sm.add_constant(X)             # add intercept
    lmodel = sm.OLS(df.cond, X)        # fit cond = a + b*temp
    results = lmodel.fit()             #
    Op = results.predict(X)            # derive 'cond' as explained by temp
    Tc1 = df.cond - Op                 # remove the linear influence

#---conditional correction --------------------------------------
    Tc = np.where(df.temp > (np.mean(df.temp) + 0.5*np.std(df.temp)), df.cond, Tc1)
    return Tc[-1]     # returning the last value

I understand it may be related to the issue:

raise NotImplementedError('See issue #11704 {url}'.format(url=url))
NotImplementedError: See issue #11704 https://github.com/pandas-dev/pandas/issues/11704

This may not have reached high enough importance for someone to look into it. Or there may be other ways we could achieve this. I am not experienced enough with pandas code to start looking into myself but I am happy try if some guidance is available.

API breaking implications

No changes in the way we call the API

Describe alternatives you've considered

Ended up using loop - much slower

[jreback]

duplicate of #15095