熊猫计算具有特定行和列的滚动总和

Question

I have quite a specific problem.

I'm trying to analyse some historic football data and want to create a couple of columns with each teams most recent goal tallys for both home and away, for example. I've tried to simplify things here, let's say the df looks like this:

df = pd.DataFrame({'Home':['A','B','C','B','A','A','C'],'Away':['B','C','A','C','B','B','A'],
                   'HG':[1,2,3,2,1,4,1],'AG':[2,4,5,1,3,2,2]})
  Home Away  HG  AG
0    A    B   1   2
1    B    C   2   4
2    C    A   3   5
3    B    C   2   1
4    A    B   1   3
5    A    B   4   2
6    C    A   1   2

What I want to do is to sum the most recent two goal numbers ( HG and/or AG ) for both Home and Away for each row in the df. But I obviously don't want to take in to account the most recent row.

So if we look at index row 0. Home is 'A'. The number I would expect to get back is 6, 5 from index row 2 under AG as A are the away team in this row, and 1 from index row 4 as A is the Home team, equaling 6 in total. For the Away team B in index row 0 I would expect the result to be 4, from index row 1 and index row 3. And so forth. I would also like to return np.NaN should it have less than 2 data points to calcualte from.

I initally thought of maybe writing a little function to help do this, something similar to this but obviously this is woefully incorrect:

def get_rolling_sum(x):
    count_list = []
    new_df = df[(df['Home'] == str(x)) | (df['Away'] == str(x))]
    for i in range(0,len(new_df)):
        if new_df['Home'].iloc[i] == str(x):
            count_list.append(new_df['HG'].iloc[i])
        elif new_df['Away'].iloc[i] == str(x):
            count_list.append(new_df['AG'].iloc[i])
df['Roll_Home'] = [get_rolling_sum(x) for x in df['Home']]

What I'm hoping to get is something like this:

  Home Away  HG  AG  Expected_Home
0    A    B   1   2            6.0
1    B    C   2   4            5.0
2    C    A   3   5            2.0
3    B    C   2   1            5.0
4    A    B   1   3            6.0
5    A    B   4   2            NaN
6    C    A   1   2            NaN

Many thanks

Answer 1

First lets add a column to the dataframe such that the row index is available. Then create a stacked dataframe so that the Home and Away columns become a single column, and the HG and AG column become a single column, while keeping the index intact. Basically the Home and Away values from the original df will become two successive rows. Then take the most recent two rows from the stacked dataframe with reference index greater than the original index and add the goals. (You have to make the last two rows NaN manually).

df = pd.DataFrame({'Home':['A','B','C','B','A','A','C'],'Away':['B','C','A','C','B','B','A'],
               'HG':[1,2,3,2,1,4,1],'AG':[2,4,5,1,3,2,2]})[['Home', 'Away', 'HG', 'AG']]
df['ref_index'] = df.index

df_stack = pd.concat([df[['Home', 'HG']].rename(columns = {'Home':'Loc', 'HG':'Goals'}), 
                  df[['Away', 'AG']].rename(columns = {'Away':'Loc', 'AG':'Goals'})]).sort_index(kind='merge')
df_stack['ref_index'] = df_stack.index

df['Expected_Home'] = df.apply(lambda row: df_stack[(df_stack.Loc == row['Home']) & 
                                                (df_stack.ref_index > row['ref_index'])].iloc[:2].Goals.sum(),
                           axis = 1)

print(df)

     Home   Away    HG  AG  ref_index   Expected_Home
   0    A      B    1   2          0    6
   1    B      C    2   4          1    5
   2    C      A    3   5          2    2
   3    B      C    2   1          3    5
   4    A      B    1   3          4    6
   5    A      B    4   2          5    2
   6    C      A    1   2          6    0