Left-align a pandas rolling object

Question

Using pandas 0.18.1, I'd like to take the rolling average of a one-column dataframe. Since version 0.18.0, this is done with rolling() objects. The default for these rolling objects is to be right-justified. There is a boolean argument you can pass, center=True, to align the rolling object to the center value, but there doesn't seem to be a way to left-align it. Here's an example:

df = pandas.DataFrame({'A': [2,3,6,8,20, 27]}) df     A 0   2 1   3 2   6 3   8 4  20 5  27 

The standard method automatically aligns to the right, so there's no value at the first two indecies with a window of size three:

 df.rolling(window=3).mean()            A 0        NaN 1        NaN 2   3.666667 3   5.666667 4  11.333333 5  18.333333 

We can center-align the operation like this:

df.rolling(window=3).mean(center=True)            A 0        NaN 1   3.666667 2   5.666667 3  11.333333 4  18.333333 5        NaN 

But what I'm looking for is this:

df.rolling(3).mean()             A  0   3.666667  1   5.666667  2  11.333333  3  18.333333  4        NaN  5        NaN 

I can accomplish this by doing it with the default right alignment, and then re-indexing it, or by reversing the order of the rows and then doing it "right-aligned" but these are work-arounds for what should be a straight-forward operation.

Answer 1

I think you can use shift:

a = df.rolling(window=3).mean().shift(-2) print (a)            A 0   3.666667 1   5.666667 2  11.333333 3  18.333333 4        NaN 5        NaN 

Answer 2

Another solution is to simply reverse the DataFrame/Series before applying the right-aligned rolling window, and re-reverse it afterwards. Something like:

In [1]: df["A"][::-1].rolling(3).mean()[::-1] Out[1]:  0     3.666667 1     5.666667 2    11.333333 3    18.333333 4          NaN 5          NaN Name: A, dtype: float64 

The benefit over shift is that it should work with variable sized windows in case of time-based windows.