0

我有一个像下面这样的数据框，它基本上是球员和他在局中得分的列表。我在这个数据框中有大约 50,000 行，数据框是根据日期排序的

PLAYER_CODE PLAYER_RUNS  MATCH_ID
123          10          1
112          5           1
123          15          2
112          10          2 
112          24          3
123          10          3
123          5           4

我需要添加新的列 PREV_TWO & PREV_THREE 这应该是他最后两局和三局的总和，从而让我跟随 DF

PLAYER_CODE PLAYER_RUNS  PREV_TWO  PREV_THREE
123          10          25        30  
112          5           34        34
123          15          15        15
112          10          24        24
112          24          0         0
123          10          5         5
123          5           0         0

我想出了以下代码来做同样的事情：

playerList = dataFrame['PLAYER_CODE'].unique().tolist()
print(len(playerList) , " Players found in dataframe")

for playerCode in playerList:
    #CREATE A PLAYER SPECIFIC DF TO LOOP AROUND THE ROWS
    playerDF = dataFrame[dataFrame['PLAYER_CODE'] == playerCode]
    playerRows = len(playerDF.index)

    i = 0
    for row in playerDF.itertuples(): #LOOP AROUND 
        j = i + 3 #TO GET THE 2-3 ROW
        x = i + 4 #TO GET THE 2-4 ROW
        #GET THE MATCH ID OF CURRENT ROW WILL BE USED TO IDENTIFY UNIQUE ROW TO UPDATE
        playerMatchId = playerDF.iloc[i]['PLAYER_MATCH_ID']
        #SUM THE WICKETS
        sumoflasttwo = playerDF.iloc[i+1:j]['PLAYER_RUNS'].sum()
        sumoflastthree = playerDF.iloc[i+1:x]['PLAYER_RUNS'].sum()

         #UPDATE THE MAIN DATA FRAME
        dataFrame.loc[(dataFrame['PLAYER_MATCH_ID'] == playerMatchId) & 
                      (dataFrame['PLAYER_CODE'] == playerCode),'LAST_TWO_AVG'] = sumoflasttwo

        dataFrame.loc[(dataFrame['PLAYER_MATCH_ID'] == playerMatchId) & 
                      (dataFrame['PLAYER_CODE'] == playerCode),'LAST_THREE_AVG'] = sumoflastthree
        i = i+1

这种方法有效，但确实很慢。我很确定有一种方法可以在没有循环的情况下做到这一点，但我不知道。有没有办法在不循环数据框的情况下做到这一点

4

1 回答 1

0

我不明白为什么PLAYER_MATCH_ID按降序排列，但PREVX列指的是下一个匹配 ID。这使得在对数据集进行故障排除时变得更加混乱，但我想通了！

首先，根据降序获取数据帧组PLAYER_MATCH_ID（以便能够使用滚动功能）。 

player_List = dataFrame['PLAYER_CODE'].unique().tolist()
groups = [dataFrame.groupby('PLAYER_CODE').get_group(i) for i in player_List]
groups = [i.sort_values('PLAYER_MATCH_ID', ascending = False) for i in groups]

接下来，使用该rolling函数获取基于PLAYER_MATCH_ID 

dfs2 = [i['PLAYER_RUNS'].rolling(min_periods = 1, window = 2).sum()
        .shift(1).fillna(0) for i in groups]
dfs3 = [i['PLAYER_RUNS'].rolling(min_periods = 1, window = 3)
        .sum().shift(1).fillna(0) for i in groups]

然后，定义一个函数来添加列（我假设您将拥有超过 2 个数据框）。

def add_columns(dat, dfs2, dfs3):
    dat['PREV_TWO'] = dfs2
    dat['PREV_THREE'] = dfs3
    return dat

最后我们可以获取最终的数据帧并将它们合并。

from functools import reduce 

final_dfs = [add_columns(groups[i], dfs2[i], 
             dfs3[i]).sort_values('PLAYER_MATCH_ID') for i in range(len(groups))]

# Final dataframe will be in order of PLAYER_CODE and sorted by PLAYER_MATCH_ID
reduce(lambda x,y: pd.merge(x,y, how = 'outer'), final_dfs)

请让我知道，如果你有任何问题。该代码可能有点棘手。

于 2019-09-16T21:13:19.800 回答