机器学习实战sklearn_随机森林

一、简介

这是一个使用天气数据集，建立随机森林模型，对数据集进行训练，从而对天气最高温度进行预测，

步骤：

1、载入数据

2、数据预处理

3、划分数据集为训练数据集，测试数据集

4、建立模型

5、预测，调优

二、具体步骤，

1、载入数据：

import pandas as pd

# 载入数据
features = pd.read_csv('data/temps.csv')
# 数据前五行
features.head(5)

# 数据特征
print('The shape of our features is:', features.shape)
features.describe()

数据前五行：

 

2、数据预处理与数据集切分：

# one-hot 编码
features = pd.get_dummies(features)
features.head(5)

# 标签与数据划分
import numpy as np

# 标签
labels = np.array(features['actual'])

# 数据
features= features.drop('actual', axis = 1)

# 转变成列表
feature_list = list(features.columns)
# 转变成np.array格式
features = np.array(features)



# 训练集与测试集划分
from sklearn.model_selection import train_test_split

train_features, test_features, train_labels, test_labels = train_test_split(features, labels, test_size = 0.25,
                                                                           random_state = 42)

 

 

3、建立随机森林模型

from sklearn.ensemble import RandomForestRegressor

# 模型建立
rf = RandomForestRegressor(n_estimators= 1000, random_state=42)
# 训练
rf.fit(train_features, train_labels)

 

 

 

4、预测

# 使用随机森林带的预测方法进行预测
predictions = rf.predict(test_features)

# 计算绝对误差
errors = abs(predictions - test_labels)
# 绝对误差，保留两位小数
print('Mean Absolute Error:', round(np.mean(errors), 2), 'degrees.')

# 计算 平均绝对百分误差mean absolute percentage error (MAPE)
mape = 100 * (errors / test_labels)

# 准确率，这里使用了平均绝对百分误差来计算准确率
accuracy = 100 - np.mean(mape)
print('Accuracy:', round(accuracy, 2), '%.')

计算结果是：

Accuracy: 93.99 %.

 

 

5、特征重要性：

# 数字特征重要性
importances = list(rf.feature_importances_)

feature_importances = [(feature, round(importance, 2)) for feature, importance in zip(feature_list, importances)]

# 排序
feature_importances = sorted(feature_importances, key = lambda x: x[1], reverse = True)

[print('Variable: {:20} Importance: {}'.format(*pair)) for pair in feature_importances];

结果：

从而可知最重要的特征是temp1,接下来我们根据此特征重新构建模型

 

6、使用最重要的特征构建模型

# 使用最重要的两个特征构建模型
rf_most_important = RandomForestRegressor(n_estimators= 1000, random_state=42)

# 获取这两个最重要的特征，划分数据集
important_indices = [feature_list.index('temp_1'), feature_list.index('average')]
train_important = train_features[:, important_indices]
test_important = test_features[:, important_indices]

# 训练
rf_most_important.fit(train_important, train_labels)

# 预测
predictions = rf_most_important.predict(test_important)

# 误差
errors = abs(predictions - test_labels)

print('Mean Absolute Error:', round(np.mean(errors), 2), 'degrees.')

mape = np.mean(100 * (errors / test_labels))
accuracy = 100 - mape

# 准确率
print('Accuracy:', round(accuracy, 2), '%.')

结果是：

Mean Absolute Error: 3.9 degrees.
Accuracy: 93.8 %.

可见，虽然使用了最重要的两个特征来构建模型，但准确率反而略微下降了一些

 

 

构建随机森林完整代码：

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor

# 载入数据,one-hot化
original_features = pd.read_csv('data/temps.csv')
original_features = pd.get_dummies(original_features)


# 标签
original_labels = np.array(original_features['actual'])

# 去除标签值
original_features= original_features.drop('actual', axis = 1)

# 特征
original_feature_list = list(original_features.columns)

# 转变成numpy.array格式
original_features = np.array(original_features)

# 切分数据集
original_train_features, original_test_features, original_train_labels, original_test_labels = train_test_split(original_features, original_labels, test_size = 0.25, random_state = 42)


# 模型建立
rf = RandomForestRegressor(n_estimators= 1000, random_state=42)

# 模型训练
rf.fit(original_train_features, original_train_labels);

# 预测
predictions = rf.predict(original_test_features)

# 计算误差
errors = abs(predictions - original_test_labels)

# 计算mean absolute error (mae)
print('Average model error:', round(np.mean(errors), 2), 'degrees.')

# 计算mean absolute percentage error (MAPE)
mape = 100 * (errors / original_test_labels)

# 计算准确率
accuracy = 100 - np.mean(mape)
print('Accuracy:', round(accuracy, 2), '%.')