我有一个通过调用Pandas.io.json.json_normalize()生成的DataFrame。这是一个
例:

dfIn = pd.DataFrame({'seed':[1324367672,1324367672,1324367673,1324367673,1324367674,1324367674], 'lanePolicy':[True,False,True,False,True,False,],
                   'stepsPerTrip':[40,37,93,72,23,70], 'density':[51,51,208,208,149,149]})


  seed       lanePolicy stepsPerTrip density
0 1324367672 True       40           51
1 1324367672 False      37           51
2 1324367673 True       93           208
3 1324367673 False      72           208
4 1324367674 True       23           149
5 1324367674 False      70           149


请注意,在dfIn['seed']中有成对的匹配值,在True中有一个False和一个dfIn['lanePolicy']值。同样,如果dfIn['seed']匹配两个给定的行,则dfIn['densitiy']也将匹配。我想计算一个类似于以下内容的表:

dfDesired = pd.DataFrame({'seed':[1324367672,1324367673,1324367674],
                   'stepsTrue':[40,93,23], 'stepsFalse':[37,72,70], 'stepsDiff':[3, 21, -47], 'density':[51,208,149]})



  seed       stepsTrue stepsFalse stepsDiff density
0 1324367672 40        37         3         51
1 1324367673 93        72         21        208
2 1324367674 23        70         -47       149


特别是,我正在寻找dfDesired['stepsDiff']中的值,这是关联的dfIn['stepsPerTrip']False和每对匹配的True对的dfIn['lanePolicy']dfIn['seed']值之间的差异。还请注意,dfDesired的行数应为dfIn的一半。

我能够使用以下方法计算该单列的值:

dfDiff = dfIn.loc[dfIn['lanePolicy'] == True]['stepsPerTrip'].reset_index()['stepsPerTrip'] - dfIn.loc[dfIn['lanePolicy'] == False]['stepsPerTrip'].reset_index()['stepsPerTrip']


0     3
1    21
2   -47
Name: stepsPerTrip, dtype: int64


但是,我想创建一个保留其他列的新DataFrame。我也尝试了以下方法,但是得到了不正确的结果:

dfDesired = dfIn.groupby('seed').apply(lambda x:x.loc[x['lanePolicy']==True]['stepsPerTrip']-x.loc[x['lanePolicy']==False]['stepsPerTrip'])


seed
1324367672  0   NaN
            1   NaN
1324367673  2   NaN
            3   NaN
1324367674  4   NaN
            5   NaN
Name: stepsPerTrip, dtype: float64


先感谢您。

最佳答案

使用DataFrame.pivot,用Series.sub减去列,对于density列,将带有seed的系列添加为与DataFrame.drop_duplicates不重复的系列:

df = dfIn.pivot('seed','lanePolicy','stepsPerTrip').add_prefix('steps')
df['stepsDiff'] = df['stepsTrue'].sub(df['stepsFalse'])
df['density'] = dfIn.drop_duplicates('seed').set_index('seed')['density']
df = df.reset_index().rename_axis(None, axis=1)
print (df)
         seed  stepsFalse  stepsTrue  stepsDiff  density
0  1324367672          37         40          3       51
1  1324367673          72         93         21      208
2  1324367674          70         23        -47      149


另一个解决方案是使用DataFrame.pivot_table和默认聚合函数mean(如果列seed,'density'和lanePolicy中有重复项):

df = (dfIn.pivot_table(index=['seed','density'], columns='lanePolicy',values='stepsPerTrip')
          .add_prefix('steps'))
df['stepsDiff'] = df['stepsTrue'].sub(df['stepsFalse'])
df = df.reset_index().rename_axis(None, axis=1)
print (df)
         seed  density  stepsFalse  stepsTrue  stepsDiff
0  1324367672       51          37         40          3
1  1324367673      208          72         93         21
2  1324367674      149          70         23        -47

10-07 21:42