将机器学习笔记 十六：基于Boruta算法的随机森林（RF）特征重要性评估与本篇结合，对比分析。

1. 探索性数据分析

输入参数： id、date、bedrooms、bathrooms、sqft_living、sqft_lot、floors、waterfront、view、condition、grade、sqft_above、sqft_basement、yr_built、yr_renovated、zipcode、lat、long、sqft_living15、sqft_lot15、
输出参数： price

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import RandomForestRegressor
from sklearn.svm import SVC
from sklearn.linear_model import SGDClassifier
from sklearn.metrics import confusion_matrix, classification_report
from sklearn.preprocessing import StandardScaler, LabelEncoder
from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score
from sklearn.model_selection import RandomizedSearchCV
from sklearn.metrics import accuracy_score
from collections import defaultdict
from sklearn.metrics import r2_score

import sys

sys.path.insert(0, 'boruta_py-master/boruta')
from boruta import BorutaPy

sys.path.insert(0, 'random-forest-importances-master/src')
from rfpimp import *

%matplotlib inline

house = pd.read_csv("C:/Users/Administrator/Desktop/kc_house_data.csv")

# 查看数据是否有空
print(house.isnull().any())
# 检查类型
print(house.dtypes)

# 删除id和date两列数据，因为他们不会使用
house = house.drop(['id', 'date'],axis=1)

用散点图展示数据之间的相关性：

with sns.plotting_context("notebook",font_scale=2.5):
    g = sns.pairplot(house[['sqft_lot','sqft_above','price','sqft_living','bedrooms']], 
                 hue='bedrooms', palette='tab20',size=6)
g.set(xticklabels=[]);

绘制参数热图（相关性分析）：

corr = house.corr()

mask = np.zeros_like(corr, dtype=np.bool)
mask[np.triu_indices_from(mask)] = True

f, ax = plt.subplots(figsize=(10, 8))
cmap = sns.diverging_palette(220, 10, as_cmap=True)

sns.heatmap(corr, mask=mask, cmap=cmap, center=0,
            square=True, linewidths=0.5)

1.1 数据集分割（训练集、测试集）

df_train, df_test = train_test_split(house, test_size=0.20,random_state=42)
df_train = df_train[list(house.columns)]
df_test = df_test[list(house.columns)]

X_train, y_train = df_train.drop('price',axis=1), df_train['price']
X_test, y_test = df_test.drop('price',axis=1), df_test['price']

X_train.shape,y_train.shape,X_test.shape,y_test.shape

((17290, 18), (17290,), (4323, 18), (4323,))

1.2 模型拟合

def predictions (rf,X_test,y_test):
    # Make predictions on test data
    predictions = rf.predict(X_test)
    # Performance metrics
    errors = abs(predictions - y_test)
    print('Metrics for Random Forest Regressor')
    print('Average absolute error:', round(np.mean(errors), 2), 'degrees.')
    # Calculate mean absolute percentage error (MAPE)
    mape = np.mean(100 * (errors / y_test))
    # Compare to baseline
    baseline_mape=np.mean(y_test)
    improvement_baseline = 100 * abs(mape - baseline_mape) / baseline_mape
    print('Improvement over baseline:', round(improvement_baseline, 2), '%.')
    # Calculate and display accuracy
    accuracy = 100 - mape
    print('Accuracy:', round(accuracy, 2), '%.')
    print('R2 score:',r2_score(predictions,y_test))

rf_reg = RandomForestRegressor(n_estimators=200,
                            min_samples_leaf=2,
                            n_jobs=-1,
                            oob_score=True,
                            random_state=42)
rf_reg.fit(X_train, y_train)

predictions(rf_reg,X_test,y_test)

Metrics for Random Forest Regressor
Average absolute error: 72704.15 degrees.
Improvement over baseline: 100.0 %.
Accuracy: 86.88 %.
R2 score: 0.8381720745711922

2. 特征重要性比较

2.1 Gini Importance

features = np.array(X_train.columns)
imps_gini=rf_reg.feature_importances_
std_gini = np.std([tree.feature_importances_ for tree in rf_reg.estimators_],
             axis=0)
indices_gini = np.argsort(imps_gini)

plt.title('Feature Importances')
plt.barh(range(len(indices_gini)), imps_gini[indices_gini], yerr=std_gini[indices_gini],color='black', align='center')
plt.yticks(range(len(indices_gini)), features[indices_gini])
plt.xlabel('Gini Importance')
plt.show()

2.2 Permutation Importance

def permutation_importances(rf, X_train, y_train, metric):
    baseline = metric(rf, X_train, y_train)
    imp = []
    std = []
    for col in X_train.columns:
        tmp=[]
        for i in range(10):
            save = X_train[col].copy()
            X_train[col] = np.random.permutation(X_train[col]) # permutation():按照给定列表生成一个打乱后的随机列表
            m = metric(rf, X_train, y_train)
            X_train[col] = save
            tmp.append(m)
        imp.append(baseline - np.mean(tmp))
        std.append(np.std(tmp))
    return np.array(imp),np.array(std)

np.random.seed(10)
imps_perm, std_perm = permutation_importances(rf_reg, X_train, y_train,oob_regression_r2_score)

features = np.array(X_train.columns)
indices_perm = np.argsort(imps_perm)

plt.title('Feature Importances')
plt.barh(range(len(indices_perm)), imps_perm[indices_perm], yerr=std_perm[indices_perm],color='black', align='center')
plt.yticks(range(len(indices_perm)), features[indices_perm])
plt.xlabel('Permutation Importance')
plt.show()

可以看出lat的重要性升高

2.3 Boruta

forest_reg = RandomForestRegressor(min_samples_leaf=2,
                                n_jobs=-1,
                                oob_score=True,
                                random_state=42)
feat_selector_reg = BorutaPy(forest_reg, verbose=2,max_iter=50)

np.random.seed(10)

import time
start = time.time()
feat_selector_reg.fit(X_train.values, y_train.values)
end = time.time()
print(end - start)

Iteration: 1 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 2 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 3 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 4 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 5 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 6 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 7 / 50
Confirmed: 0
Tentative: 18
Rejected: 0
Iteration: 8 / 50
Confirmed: 13
Tentative: 0
Rejected: 5

BorutaPy finished running.
Iteration: 9 / 50
Confirmed: 13
Tentative: 0
Rejected: 5
837.3257942199707

print('Confirmed: \n',list(np.array(X_train.columns)[feat_selector_reg.ranking_==1]))
print('\nTentatives: \n',list(np.array(X_train.columns)[feat_selector_reg.ranking_==2]))
print('\nRejected: \n',list(np.array(X_train.columns)[feat_selector_reg.ranking_>=3]))

Confirmed:
[‘bathrooms’, ‘sqft_living’, ‘sqft_lot’, ‘waterfront’, ‘view’, ‘grade’, ‘sqft_above’, ‘yr_built’, ‘zipcode’, ‘lat’, ‘long’, ‘sqft_living15’, ‘sqft_lot15’]

Tentatives:
[‘sqft_basement’]

Rejected:
[‘bedrooms’, ‘floors’, ‘condition’, ‘yr_renovated’]

---

3. 特征比较

3.1 Gini Importance

X_train_gini_reg=X_train[['grade','sqft_living','lat','long']]
X_test_gini_reg=X_test[['grade','sqft_living','lat','long']]

rf_gini_reg = RandomForestRegressor(n_estimators=200,
                            min_samples_leaf=2,
                            n_jobs=-1,
                            oob_score=True,
                            random_state=42)
rf_gini_reg.fit(X_train_gini_reg, y_train)

3.2 Permutation Importance

X_train_perm_reg=X_train.drop(['bedrooms','yr_renovated','floors','sqft_basement','condition','bathrooms'],axis=1)
X_test_perm_reg=X_test.drop(['bedrooms','yr_renovated','floors','sqft_basement','condition','bathrooms'],axis=1)

rf_perm_reg = RandomForestRegressor(n_estimators=200,
                            min_samples_leaf=2,
                            n_jobs=-1,
                            oob_score=True,
                            random_state=42)
rf_perm_reg.fit(X_train_perm_reg, y_train)

3.3 Boruta

X_train_boruta_reg=X_train.drop(['bedrooms','floors','condition','yr_renovated'],axis=1)
X_test_boruta_reg=X_test.drop(['bedrooms','floors','condition','yr_renovated'],axis=1)

rf_boruta_reg = RandomForestRegressor(n_estimators=200,
                            min_samples_leaf=2,
                            n_jobs=-1,
                            oob_score=True,
                            random_state=42)
rf_boruta_reg.fit(X_train_boruta_reg, y_train)

4. 模型比较

print('******************* Original Model ***********************')
print('\n')
predictions(rf_reg,X_test,y_test)

print ('\n')

print('**** Feature selection based on Gini Importance ****')
print('\n')
predictions(rf_gini_reg,X_test_gini_reg,y_test)

print ('\n')

print('**** Feature selection based on Permutation Importance *****')
print('\n')
predictions(rf_perm_reg,X_test_perm_reg,y_test)

print ('\n')

print('*********** Feature selection based on Boruta **************')
print('\n')
predictions(rf_boruta_reg,X_test_boruta_reg,y_test)

******************* Original Model ***********************

Metrics for Random Forest Regressor
Average absolute error: 72704.15 degrees.
Improvement over baseline: 100.0 %.
Accuracy: 86.88 %.
R2 score: 0.8381720745711922

**** Feature selection based on Gini Importance ****

Metrics for Random Forest Regressor
Average absolute error: 81288.41 degrees.
Improvement over baseline: 100.0 %.
Accuracy: 85.56 %.
R2 score: 0.8052584664901095

**** Feature selection based on Permutation Importance *****

Metrics for Random Forest Regressor
Average absolute error: 72741.67 degrees.
Improvement over baseline: 100.0 %.
Accuracy: 86.77 %.
R2 score: 0.8477802122659206

*********** Feature selection based on Boruta **************

Metrics for Random Forest Regressor
Average absolute error: 73254.05 degrees.
Improvement over baseline: 100.0 %.
Accuracy: 86.75 %.
R2 score: 0.8388239891237698

Permutation Importance对于R2的计算是比较好的模型，Permutation Importance和Boruta都是比较好的方法。