如何在Python的数据框中创建嵌套列？

Question

我有三列数据，分别为模型（应该作为索引），不使用标准化的准确率，以及采用标准化的准确率（使用zscore、minmax、maxabs、robust），需要按照以下格式创建：

 ------------------------------------------------------------------------------------
|   Models  |  Accuracy without normalization    |      Accuracy with normalization  |
|           |                                    |-----------------------------------|
|           |                                    | zscore | minmax | maxabs | robust |
 ------------------------------------------------------------------------------------

dfmod-> Models column
dfacc-> Accuracy without normalization
dfacc1-> Accuracy with normalization - zscore
dfacc2-> Accuracy with normalization - minmax
dfacc3-> Accuracy with normalization - maxabs
dfacc4-> Accuracy with normalization - robust

dfout=pd.DataFrame({('Accuracy without Normalization'):{dfacc},
     ('Accuracy using Normalization','zscore'):{dfacc1},
     ('Accuracy using Normalization','minmax'):{dfacc2},
     ('Accuracy using Normalization','maxabs'):{dfacc3},
     ('Accuracy using Normalization','robust'):{dfacc4},
   },index=dfmod
)

我想尝试像这样做，但我无法进一步理解。

测试数据：

qda    0.6333       0.6917      0.5917      0.6417     0.5833
svm    0.5333       0.6917      0.5333      0.575      0.575
lda    0.5333       0.6583      0.5333      0.5667     0.5667
lr     0.5333       0.65        0.4917      0.5667     0.5667
dt     0.5333       0.65        0.4917      0.5667     0.5667
rc     0.5083       0.6333      0.4917      0.525      0.525
nb     0.5          0.625       0.475       0.5        0.4833
rfc    0.5          0.625       0.4417      0.4917     0.4583
knn    0.3917       0.6         0.4417      0.4833     0.45
et     0.375        0.5333      0.4333      0.4667     0.45
dc     0.375        0.5333      0.4333      0.4667     0.425
qds    0.3417       0.5333      0.4         0.4583     0.3667
lgt    0.3417       0.525       0.3917      0.45       0.3583
lt     0.2333       0.45        0.3917      0.4167     0.3417

这些是按照上述表格中指定的顺序对应的各子列的值。

- itti_da

---

插入一些测试数据。 - hpchavaz

Answer 1

有一个比较糟糕的方法可以做到这一点，我会写下来，直到有人用更好的办法回答。我们开始吧：

import pandas as pd

# I assume that you can read raw data named test.csv by pandas and
# set header = None cause you mentioned the Test data without any headers, so:
df = pd.read_csv("test.csv", header = None)

# Then define preferred Columns! 
MyColumns = pd.MultiIndex.from_tuples([("Models" , ""),
                                       ("Accuracy without normalization" , ""),
                                       ("Accuracy with normalization" , "zscore"),
                                       ("Accuracy with normalization" , "minmax"),
                                       ("Accuracy with normalization" , "maxabs"),
                                       ("Accuracy with normalization" , "robust")])

# Create new DataFrame with specified Columns, after this you should pass values 
New_DataFrame = pd.DataFrame(df , columns = MyColumns)

# a loop for passing values
for item in range(len(MyColumns)):
    New_DataFrame.loc[: , MyColumns[item]] = df.iloc[: , item]

这给了我：

在将New_DataFrame的索引设置为Models时，你可以继续执行：

New_DataFrame.set_index(New_DataFrame.columns[0][0] , inplace=True)
New_DataFrame

这给了我：

- Shayan