Python Pandas：子类系列和数据框

Question

I want to add new methods and attributes to pandas Series and DataFrames . Here is very simplistic example: I want a method that counts the number of time the difference between a row and the previous one is not 1.

Here is what I had so far by sub-classing the pandas objects:

import pandas as pd

class Serie(pd.Series):

    def gaps(self):
        return (self.diff().fillna(1) != 1).sum()

class DataSet(pd.DataFrame):

    _constructor_sliced = Serie

But based on this answer it seems that I can do this instead:

def gaps(self):
    return (self.diff().fillna(1) != 1).sum()

pd.Series.gaps = gaps

It seems to work equally well!

In[1]: df = pd.DataFrame({'A':[1,2,4], 'B':[3,2,1]})
In[2]: df.A.gaps()
Out[2]: 1

Now the question: what is the best practice for such situations? Second option seems much simpler than subclassing but I may be missing something... Are there caveats from doing that? Or maybe there are other options I missed.

Answer 1

A very simple solution would be too just use a function like your gaps function but renaming "self" to "serie":

def gaps(serie):
    return (serie.diff().fillna(1) != 1).sum()

It keeps concerns cleanly separated (your code vs pandas code).

It's more readable: You don't need to understand a lot of things to understand how it works. It's just simple.

It's less surprising: A dev on your team may spend some time trying to search for the gap() documentation in the Pandas' Series documentation and not finding it, just to discover a few hours later that someone (you) monkey patched it.

It's also the shortest solution.

It avoids using "private" members like _constructor_sliced which name may change in the future and break your implementation.

It avoids future conflicts: What about the next release of pandas include a gap method in the Series object? It won't directly break, but I'll bit a developper in your team, wanting to use the "now well known .gap()", not being aware you "changed it", and having hard times debugging why it does not work according to the documentation.