熊猫 ffill 和 bfill 不能一起工作

Question

I am having trouble making bfill and ffill work within the same dataset.

I have a merged dataset similar to the one below. All rows have a Project Code and Date, but rows where spending was recorded before/after the Start/End Date of a subscription do not have a subscription code.

Project Code     Start Date     End Date     Subscription Code     Date     Recorded Spending 
   349                                                            8/1/19          50
   349             9/1/18        9/1/19          349A             3/1/19          88
   349             9/1/18        9/1/19          349A             8/1/19          
   349             9/1/19        9/1/20          349B             10/1/19         120
   349                                                            10/1/20         22

I would like to extend the Subscription Code values so that all spending before the official start of the project is counted under the first subscription code and any spending after the official completion of the project is counted under the last subscription code.

In my solution I have found that I can EITHER ffill or bfill - whichever is first. So the code below results in forward-filled Subscription Codes but codes are never back filled.

    df.sort_values(by=['Project Code','Date'], inplace=True)
    #backfill subscription code
    df.loc[:,['Subscription Code']] = df.loc[:,['Subscription Code']].ffill()
    
    #remove if the project code does not match subscription code
    df['Subscription Code'] = np.where(df['Subscription Code'].str[:3] != df['Project Code'], '', df['Subscription Code'])

    df.loc[:,['Subscription Code']] = df.loc[:,['Subscription Code']].bfill()

    #remove if the project code does not match subscription code
    df['Subscription Code'] = np.where(df['Subscription Code'].str[:3] != df['Project Code'], '', df['Subscription Code'])

How to I combine these so that I can BOTH ffill and bfill?

Answer 1

In this case I knew that the first subscription code would be the Project Code + 'A', so I was able to use this code to get what I needed:

def fill_empty_subscription_code(df):
    
    df.sort_values(by=['Project Code', 'Date'], inplace=True) 
    #If there is spending recorded after the last subscription end date then include it with the last subscription
    df.loc[:, 'Subscription Code'] = df.groupby(['Project Code'])['Subscription Code'].ffill()
    #If there is work done before the first subscription start date include it in the first subscription line code (denoted with the concatenated  0)
    df.loc[:, 'Subscription Code'] = np.where(pd.isna(df['Subscription Code']), 
      df['Project Code'] + 'A',
      df['Subscription Code'])
            
    return df