“特征重要性”的“单热编码”变量的显示名称

Question

在完成算法的训练和验证后，如何正确显示“one-hot-encoded”特征的名称？我想整齐地显示每个功能的名称及其重要性。以下是我尝试过的：

显示特征重要性：

grid_search.best_estimator_.feature_importances_
array([  7.67359589e-02,   7.20731884e-02,   4.38667330e-02,
         1.69222269e-02,   1.51816327e-02,   1.66947835e-02,
         1.56858183e-02,   3.43347923e-01,   5.95555727e-02,
         7.65422356e-02,   1.11224727e-01,   1.02677088e-02,
         1.32720377e-01,   1.06447326e-04,   4.45207929e-03,
         4.62258699e-03])

获取热门类别名称：

cat_one_hot_attribs = list(encoder.classes_)
print(cat_one_hot_attribs)
['<1H OCEAN', 'INLAND', 'ISLAND', 'NEAR BAY', 'NEAR OCEAN']

获取其余名称（其他类别）：

num_attribs = list(X_train)

['longitude',
 'latitude',
 'housing_median_age',
 'total_rooms',
 'total_bedrooms',
 'population',
 'households',
 'median_income',
 'rooms_per_household',
 'bedrooms_per_household',
 'population_per_household',
 0,
 1,
 2,
 3,
 4]

现在我执行以下操作：

attributes = num_attribs + cat_one_hot_attribs

print(pd.DataFrame(sorted(zip(feature_importance, attributes), reverse=True)))

但我得到以下信息：

         0                         1
0   0.343348             median_income
1   0.132720                         1
2   0.111225  population_per_household
3   0.076736                 longitude
4   0.076542    bedrooms_per_household
5   0.072073                  latitude
6   0.059556       rooms_per_household
7   0.043867        housing_median_age
8   0.016922               total_rooms
9   0.016695                population
10  0.015686                households
11  0.015182            total_bedrooms
12  0.010268                         0
13  0.004623                         4
14  0.004452                         3
15  0.000106                         2

我也尝试了其他方法，但都失败了。

有人可以建议一种正确显示此内容的方法吗？谢谢。

编辑：

根据@cᴏʟᴅsᴘᴇᴇᴅ 的回答，我尝试了以下方法：

feature_importance = grid_search.best_estimator_.feature_importances_

cat_one_hot_attribs = list(encoder.classes_)

num_attribs = list(X_train)
attributes = num_attribs + cat_one_hot_attribs

vals = sorted(zip(feature_importance, attributes), key=lambda x: x[0], reverse=True)
df = pd.DataFrame(vals)
print(df)

仍然得到如上的输出。

Answer 1

打破它。首先排序，按键。确保只考虑 feature_importances。

设置：

import pandas as pd
import numpy as np

feature_importance = np.array([  7.67359589e-02,   7.20731884e-02,   4.38667330e-02,
     1.69222269e-02,   1.51816327e-02,   1.66947835e-02,
     1.56858183e-02,   3.43347923e-01,   5.95555727e-02,
     7.65422356e-02,   1.11224727e-01,   1.02677088e-02,
     1.32720377e-01,   1.06447326e-04,   4.45207929e-03,
     4.62258699e-03])

cat_one_hot_attribs = ['<1H OCEAN', 'INLAND', 'ISLAND', 'NEAR BAY', 'NEAR OCEAN']

num_attribs = ['longitude',
 'latitude',
 'housing_median_age',
 'total_rooms',
 'total_bedrooms',
 'population',
 'households',
 'median_income',
 'rooms_per_household',
 'bedrooms_per_household',
 'population_per_household',
 0,
 1,
 2,
 3,
 4]

attributes = num_attribs

---

通过feature_importance 获取vals 的排序列表。

vals = sorted(zip(feature_importance, attributes), key=lambda x: x[0], reverse=True)
df = pd.DataFrame(vals)

然后，使用.replace 将编码替换为cat_one_hot_attribs 中的值。

df.iloc[:, -1] = df.iloc[:, -1].replace({i : k for i, k in enumerate(cat_one_hot_attribs)})
df

           0                         1
0   0.343348             median_income
1   0.132720                    INLAND
2   0.111225  population_per_household
3   0.076736                 longitude
4   0.076542    bedrooms_per_household
5   0.072073                  latitude
6   0.059556       rooms_per_household
7   0.043867        housing_median_age
8   0.016922               total_rooms
9   0.016695                population
10  0.015686                households
11  0.015182            total_bedrooms
12  0.010268                 <1H OCEAN
13  0.004623                NEAR OCEAN
14  0.004452                  NEAR BAY
15  0.000106                    ISLAND