Question

我可以看到用于垂直排序记录的解决方案，但是我想水平排列数据框中的数据子集。

这是我的数据框，其中包含我要排序的数据：

account_num Word_0    Word_1    Word_2    Word_3    Word_4
123         Silver    Platinum  Osmium    
456         Platinum  
789         Silver    Rhodium   Platinum  Osmium

这是我想要的输出：

account_num  Word_0     Word_1    Word_2   Word_3   Word_4
123          Platinum   Osmium    Silver   
456          Platinum   
789          Rhodium    Platinum  Osmium   Silver

根据此数据框内的顺序：

Priority    Metal
1           Rhodium
2           Platinum
3           Gold
4           Ruthenium
5           Iridium
6           Osmium
7           Palladium
8           Rhenium
9           Silver
10          Indium

我已经使用这段代码整理了数据：

newdf.apply(lambda r: sorted(r,reverse = True), axis = 1)

其中将Word_0至4列放置在另一个数据帧（newdf）中，然后以相反的顺序排序，因此最后出现空白值，然后将它们重新连接到包含account_num列的原始数据帧中，但是我不知道如何合并订购顺序中的自定义列表。

任何帮助将不胜感激

谢谢

Answer 1

我觉得我们可以melt，merge顺序df，然后sort_values基于Priority，然后pivot返回

s=df.melt('account_num').\
     merge(orderdf,left_on='value',right_on='Metal',how='left').\
       sort_values('Priority')
yourdf=s.assign(newkey=s.groupby('account_num').cumcount()).\
           pivot('account_num','newkey','value').add_prefix('Word_')
yourdf
Out[1100]: 
newkey         Word_0    Word_1  Word_2  Word_3 Word_4
account_num                                           
123          Platinum    Osmium  Silver    None    NaN
456          Platinum      None    None    None    NaN
789           Rhodium  Platinum  Osmium  Silver    NaN

或者我们对argsort使用更清晰的逻辑

d = dict(zip(df2['Metal'], df2['Priority']))
for x in range(len(df)):

    df.iloc[x,:]=df.values[x,np.argsort([d.get(x) if x ==x else 1000 for x in df.values[x,:]] )]

df
Out[38]: 
                 Word_0    Word_1  Word_2  Word_3  Word_4
  account_num                                            
0 123          Platinum    Osmium  Silver     NaN     NaN
1 456          Platinum       NaN     NaN     NaN     NaN
2 789           Rhodium  Platinum  Osmium  Silver     NaN

Answer 2

使用pd.Categorical

c = pd.Categorical(df2.Metal, df2.Metal, ordered=True)

df.set_index('account_num').transform(lambda k: pd.Categorical(k, 
                                                           categories=c.categories)\
                                  .sort_values(), axis=1)

输出

            Word_0       Word_1     Word_2  Word_3  Word_4
account_num                 
123         Platinum     Osmium     Silver  NaN     NaN
456         Platinum     NaN        NaN     NaN     NaN
789         Rhodium      Platinum   Osmium  Silver  NaN

当然，总是可以.fillna('')结尾。

Answer 3

您也可以尝试：

df=df.fillna(value=pd.np.nan)
d=dict(zip(ref.Metal,ref.Priority))
df[['account_num']].join(pd.DataFrame(np.sort(df.iloc[:,1:].replace(d).values,axis=1),
                        columns=df.iloc[:,1:].columns).replace({v:k for k,v in d.items()}))

   account_num    Word_0    Word_1  Word_2  Word_3 Word_4
0          123  Platinum    Osmium  Silver     NaN    NaN
1          456  Platinum       NaN     NaN     NaN    NaN
2          789   Rhodium  Platinum  Osmium  Silver    NaN

Answer 4

使用：

#create helper dictionary
d = dict(zip(df2['Metal'], df2['Priority']))
#add empty string for maximum priority
d[''] = df2['Priority'].max() + 1

#use sorted by key and dictioanry
L = [sorted(x, key=d.get) for x in df.fillna('').values]
#create new DataFrame by constructor
df1 = pd.DataFrame(L, index=df.index).add_prefix('Word_')
print (df1)
               Word_0    Word_1  Word_2  Word_3 Word_4
account_num                                           
123          Platinum    Osmium  Silver               
456          Platinum                                 
789           Rhodium  Platinum  Osmium  Silver

如果需要缺少值：

df1 = pd.DataFrame(L, index=df.index).add_prefix('Word_').replace('', np.nan)
print (df1)
               Word_0    Word_1  Word_2  Word_3  Word_4
account_num                                            
123          Platinum    Osmium  Silver     NaN     NaN
456          Platinum       NaN     NaN     NaN     NaN
789           Rhodium  Platinum  Osmium  Silver     NaN

根据自定义列表跨数据框的列排序

4 个答案: