熊猫。应用扩展列值错误：如果使用所有标量值，则必须传递索引

Question

我想对数据帧应用一个函数，该数据帧为原始数据集中的每一列返回几列。apply函数返回包含列和索引的DataFrame，但仍会引发错误ValueError：如果使用所有标量值，则必须传递索引。

我试着设置输出数据帧的名称，将列设置为多索引，并将索引设置为多索引，但都不起作用。

示例：我有这个输入数据帧

df_all_users = pd.DataFrame(
    [[1, 2, 3],
     [1, 2, 3],
     [1, 2, 3],
    ],
    index=["2020-01-01", "2020-01-02", "2020-01-03"],
    columns=["user_1", "user_2", "user_3"])

          user_1  user_2    user_3
2020-01-01     1       2         3
2020-01-02     1       2         3
2020-01-03     1       2         3

apply_函数如下所示：

def apply_function(df):
    df_out = pd.DataFrame(index=df.index)
    # these columns are in reality computed used some other functions
    df_out["column_1"] = df.values  # example: pyod.ocsvm.OCSVM.fit_predict(df.values) 
    df_out["column_2"] = - df.values  # example: pyod.knn.KNN.fit_predict(df.values)
    
    # these are the things I've tried without working
    df_out.name = df.name
    df_out.columns = pd.MultiIndex.from_tuples([(df.name, column) for column in df_out.columns],
                                               names=["user", "score"])
    df_out.index = pd.MultiIndex.from_tuples([(df.name, idx) for idx in df_out.index],
                                             names=["user", "date"])
    print(df_out)
    return df_out

df_all_users.apply(apply_function, axis=0, result_type="expand")

这会引发错误：

ValueError：如果使用所有标量值，则必须传递索引

我期望的输出如下所示：

out_df = pd.DataFrame(
    [[1, 1, 2, 2, 3, 3],
     [1, 1, 2, 2, 3, 3],
     [1, 1, 2, 2, 3, 3],
    ],
    index=["2020-01-01", "2020-01-02", "2020-01-03"],
    columns=pd.MultiIndex.from_tuples([(user, column)
                                       for user in ["user_1", "user_2", "user_3"]
                                       for column in ["column_1", "column_2"]],
                                      names=("user", "score"))
)

             user_1           user_2            user_3
           column_1 column_2 column_1 column_2 column_1 column_2
2020-01-01        1        1        2        2        3        3
2020-01-02        1        1        2        2        3        3
2020-01-03        1        1        2        2        3        3

Answer 1

好的，答案是将输出转换为一系列数组，然后连接结果：

import pandas as pd
df_all_users = pd.DataFrame(
    [[1, 2, 3],
     [1, 2, 3],
     [1, 2, 3],
    ],
    index=["2020-01-01", "2020-01-02", "2020-01-03"],
    columns=["user_1", "user_2", "user_3"])

def apply_function(df):
    df_out = pd.DataFrame(index=df.index)
    df_out["column_1"] = df.values
    df_out["column_2"] = df.values

    df_out = pd.Series([values for values in df_out.values], index=df.index)
    df_out.name = df.name
    return df_out

df_out = df_all_users.groupby(level=0, axis=1).apply(apply_function)
df_out = pd.DataFrame([np.concatenate(values, axis=0) for values in df_out.values],
                      index=df_out.index,
                      columns=pd.MultiIndex.from_tuples([(user, column)
                                                         for column in ["column_1", "column_2"]
                                                         for user in df_out.columns
                                                        ], names=["user", "algorithm"]))
df_out



user          user_1                  user_2                  user_3
algorithm   column_1    column_2    column_1    column_2    column_1    column_2
2020-01-01         1           1           2           2           3           3
2020-01-02         1           1           2           2           3           3
2020-01-03         1           1           2           2           3           3

Answer 2

这样做：

import numpy as np

df_all_users[np.repeat(df_all_users.columns.values,2)]