DHU数据科学技术与应用 第八次作业——机器学习建模分析

作业——机器学习建模分析"/>

DHU数据科学技术与应用 第八次作业——机器学习建模分析

全部都是啥也不会的薯条自己抄书+网上查资料瞎写的，不保证是正确答案。

不要照搬！不要照搬！不要照搬！

1. (简答题, 50分)

Energy Efficiency数据集（ ENB2012_data.xlsx，ENB2012.names）记录不同房屋的制热能源消耗和制冷能源消耗。包括768条记录，8个特征属性，两个预测值。具体说明见ENB2012.names。

1. 1）在全数据集上训练线性回归模型预测制热能耗，计算模型性能：RMSE以及R2；

2. 2）将数据集划分训练集和测试集，在训练集上训练线性回归模型，分析模型在训练集

和测试集上的性能。

import pandas as pd
from sklearn import model_selection
from sklearn import metrics
from sklearn.linear_model import LinearRegression# 1) 在全数据集上训练线性回归模型预测制热能耗，计算模型性能：RMSE以及R2
filename = 'data\ENB2012_data.xlsx'
data = pd.read_excel(filename)
# print(data.iloc[0:769,:].values)X = data.iloc[:,0:8]
y = data.iloc[:,8]    #制热
linreg = LinearRegression()
linreg.fit(X,y)
print("intercept: ",linreg.intercept_,"\ncoefficient: ",linreg.coef_)  #截距+回归系数y_pred = linreg.predict(X)
err = metrics.mean_squared_error(y,y_pred)    # RMSE
print('The mean square error is: {:2f}'.format(err))
predict_score1 = linreg.score(X,y)   # R^2
print('The decision coefficient is: {:2f}'.format(predict_score1),"\n")# 2) 将数据集划分训练集和测试集，在训练集上训练线性回归模型，分析模型在训练集和测试集上的性能
y = data.iloc[:,8:10]    #制热+制冷
X_train,X_test,y_train,y_test = model_selection.train_test_split(X,y,test_size=0.35,random_state=1)     #切分linregTr = LinearRegression()
linregTr.fit(X_train,y_train)
print("intercept: ",linregTr.intercept_,"\ncoefficient: ",linregTr.coef_)  #截距+回归系数y_train_pred = linregTr.predict(X_train)
y_test_pred = linregTr.predict(X_test)
train_err = metrics.mean_squared_error(y_train,y_train_pred)    # RMSE
test_err = metrics.mean_squared_error(y_test,y_test_pred)   # RMSE
print('The mean square error of train and test are: {:2f},{:2f}'.format(train_err,test_err))
predict_score2 = linregTr.score(X_test,y_test)   # R^2
print('The decision coefficient is: {:2f}'.format(predict_score2))

2. (简答题, 50分)

基于bankpep数据集，划分训练集与测试集，建立分类模型。 

1. 1）使用决策树在训练集上建立分类模型，记录模型在测试集上的性能；

2. 2）自学朴素贝叶斯、支持向量集训练分类模型的方法，分别在训练集上建立分类模型，记录模型在测试集上的性能；

3. 3) 分析比较三种方法的性能差异。

1）决策树

# 基于bankpep数据集，划分训练集与测试集，建立分类模型import pandas as pd
from sklearn import model_selection, treefilename = 'data\bankpep.csv'
data1 = pd.read_csv(filename, index_col=0, header=None, skiprows=1)
# print(data)
data1.loc[data1[2] == 'FEMALE', 2] = 1
data1.loc[data1[2] == 'MALE', 2] = 0
data1.loc[data1[3] == 'INNER_CITY', 3] = 0
data1.loc[data1[3] == 'TOWN', 3] = 1
data1.loc[data1[3] == 'RURAL', 3] = 2
data1.loc[data1[3] == 'SUBURBAN', 3] = 3
data1.loc[data1[5] == 'YES', 5] = 1
data1.loc[data1[5] == 'NO', 5] = 0
data1.loc[data1[6] == 'YES', 6] = 1
data1.loc[data1[6] == 'NO', 6] = 0
data1.loc[data1[7] == 'YES', 7] = 1
data1.loc[data1[7] == 'NO', 7] = 0
data1.loc[data1[8] == 'YES', 8] = 1
data1.loc[data1[8] == 'NO', 8] = 0
data1.loc[data1[9] == 'YES', 9] = 1
data1.loc[data1[9] == 'NO', 9] = 0
data1.loc[data1[10] == 'YES', 10] = 1
data1.loc[data1[10] == 'NO', 10] = 0
data1.loc[data1[11] == 'YES', 11] = 1
data1.loc[data1[11] == 'NO', 11] = 0
# print(data, '\n')# 1) 使用决策树在训练集上建立分类模型，记录模型在测试集上的性能
X1 = data1.iloc[:, 0:10]
y1 = data1.iloc[:, 10]
# linreg = LinearRegression()
# linreg.fit(X,y)
# print(linreg.intercept_,linreg.coef_)
X_train1, X_test1, y_train1, y_test1 = model_selection.train_test_split(X1, y1, test_size=0.35, random_state=1)
clf = tree.DecisionTreeClassifier()
clf = clf.fit(X_train1, y_train1.astype('float64'))
score1 = clf.score(X_test1, y_test1.astype('float64'))
print('1) 决策树\nscore =', score1)

2） 朴素贝叶斯

资料来源：朴素贝叶斯算法（python 实现）

在这里需要注意：上方资料中的贝叶斯算法使用数据集类型数据进行计算，而先前题目中data数据是dataframe类型。薯条在这里尝试了一个下午，最后选择了强行转换为数组形式，等效于数据集代入公式计算。

# 2) 自学朴素贝叶斯、支持向量机训练分类模型的方法，分别在训练集上建立分类模型，记录模型在测试集上的性能
# 朴素贝叶斯
import pandas as pd
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from collections import Counter
import time# 创建矩阵装载dataframe数据，强制转换为数据集类型
data2 = data1
train2 = data2.drop([11], axis=1).astype('float')
test2 = data2[11].astype('float')
train_array = train2.values
test_array = test2.values
train_array.reshape(600, 10)
test_array.reshape(600, 1)# print(train_array)class Beyes(object):def __init__(self):self.length = -1  # 保存测试集数据量self.train_target_list = []  # 目标值类别集合self.p_train_target = {}  # 保存个目标值概率self.split_data_lis = []  # 保存各条件概率对应的数据集self.feature_p_lis = []  # 保存特征概率self.predict = []  # 保存分类结果def fit(self, train_data, train_target):"""P(X)=∑kP(X|Y=Yk)P(Yk)计算P(Yk)：p_train_target准备计算条件概率所需数据集：self.split_data_lis:param train_data:<arr_like>:param train_target:<arr_like>:return:"""train_length = train_data.shape[0]self.length = train_lengthtarget_list = list(set(train_target))  # 队训练集目标值去重self.train_target_list = target_list  # 写入对象特征target_classifier = dict(Counter(train_target))  # 保存目标值的分类计数（字典格式）train_data = pd.DataFrame(train_data)train_data['target'] = train_target  # 将数据转换为DataFrame格式方便后续聚合for target in self.train_target_list:self.p_train_target[target] = target_classifier[target] / self.length  # 保存各目标值的概率split_data = train_data[train_data['target'] == target]self.split_data_lis.append(split_data)print('model had trained please use classifier() to get result')def p_test_data(self, sample):"""映射函数：取测试集的一个样本，计算各特征对应条件概率及最终的分类结果并将结果映射到测试集的target列:param sample:Serise:return self.train_target_list[position]:概率最大的类别"""result_p = []for j in range(len(self.train_target_list)):p_label = 1this_target = self.train_target_list[j]this_data = self.split_data_lis[j]for i in range(0, sample.shape[0]):feature_num_dict = dict(Counter(this_data[i]))  # 计算一列数据中各类别的数量if sample[i] in feature_num_dict:label_num = feature_num_dict.get(sample[i])p_label = p_label * (label_num / this_data.shape[0])  # 计算单个特征的条件概率else:# 加入拉普拉斯平滑系数解决概率为0的情况'p_label = p_label * (1 / (this_data.shape[0] + len(feature_num_dict)))this_target_p = p_label * self.p_train_target.get(this_target)  # 计算该样本属于该特征的概率result_p.append(this_target_p)position = result_p.index(max(result_p))  # 概率最大的分类return self.train_target_list[position]def classifier(self, test_data):"""计算分类结果:param test_data:Serise:return:"""if self.length == -1:raise ValueError('please use fit() to train the train data set ')else:test_data = pd.DataFrame(test_data)test_data['target'] = test_data.apply(self.p_test_data, axis=1)  #self.predict = list(test_data['target'])print('classfier result:', self.predict)def score(self, test_target):"""计算准确率:param test_target::return:"""if len(self.predict) == 0:raise ValueError('please use classifier() to get classifier target')else:count = 0for i in range(0, test_target.shape[0]):if test_target[i] == self.predict[i]:count += 1score = count / (test_target.shape[0])print('the Classification accuracy is:', score)if __name__ == '__main__':""";用sk_learn中的鸢尾花数据集测试"""# iris = load_iris()x = train_array  # iris.datay = test_array  # iris.target# print(x,y) x为二维矩阵、y为一维矩阵 np.arrayprint('\n2) 朴素贝叶斯')x_train, x_test, y_train, y_test = train_test_split(x, y)print('训练集数据量', x_train.shape)print('测试集数据量', x_test.shape)start_time = time.time()classifier = Beyes()classifier.fit(x_train, y_train)classifier.classifier(x_test)classifier.score(y_test)end_time = time.time()time_d = end_time - start_timeprint("spend time:", time_d)

 3）支持向量机

# 支持向量机
data3 = pd.read_csv(filename, index_col='id')
print('\n3) 支持向量机\n')  # ,data3)
seq = ['married', 'car', 'save_act', 'current_act', 'mortgage', 'pep']
for feature in seq:data3.loc[data3[feature] == 'YES', feature] = 1data3.loc[data3[feature] == 'NO', feature] = 0
data3.loc[data3['sex'] == 'FEMALE', 'sex'] = 1
data3.loc[data3['sex'] == 'MALE', 'sex'] = 0
# print(data3)dumm_reg = pd.get_dummies(data3['region'], prefix='region')
dumm_child = pd.get_dummies(data3['children'], prefix='children')
df1 = data3.drop(['region', 'children'], axis=1)
df2 = df1.join([dumm_reg, dumm_child], how='outer')
# print(df2[0:5])X3 = df2.drop(['pep'], axis=1).astype('int')
y3 = df2['pep'].astype('int')
# print('X3\n',X3, '\ny3\n',y3)from sklearn import svmclf = svm.SVC(kernel='rbf', gamma=0.6, C=100)
clf.fit(X3, y3)
print("Accuracy:", clf.score(X3, y3))from sklearn import metricsy3_predicted = clf.predict(X3)
print(metrics.classification_report(y3, y3_predicted))X_train3, X_test3, y_train3, y_test3 = model_selection.train_test_split(X3, y3, test_size=0.3, random_state=1)
clf = svm.SVC(kernel='rbf', gamma=0.6, C=0.001)
clf.fit(X_train3, y_train3)
print("Performance on training set:", clf.score(X_train3, y_train3))
print("Performance on test set:", clf.score(X_test3, y_test3))from sklearn import preprocessingX_scale3 = preprocessing.scale(X3)
X_train3, X_test3, y_train3, y_test3 = model_selection.train_test_split(X_scale3, y3, test_size=0.3, )
clf = svm.SVC(kernel='poly', gamma=0.6, C=0.001)
clf.fit(X_train3, y_train3)
score3 = clf.score(X_test3, y_test3)
print('\nscore =', score3)

第二题三个代码为同一文件。

hxd们发现错误一定要戳我！

补充：仅供参考，注意查重，禁止照搬