机器学习十讲--第三讲-分类

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

import pandas as pd

raw_train = pd.read_csv("input/chinese_news_cutted_train_utf8.csv",sep="	",encoding="utf8")
raw_test = pd.read_csv("input/chinese_news_cutted_test_utf8.csv",sep="	",encoding="utf8")

raw_train_binary = raw_train[((raw_train["分类"] == "科技") | (raw_train["分类"] == "文化"))]
raw_test_binary = raw_test[((raw_test["分类"] == "科技") | (raw_test["分类"] == "文化"))]

stop_words = []
file = open("input/stopwords.txt",encoding='UTF-8')
for line in file:
    stop_words.append(line.strip())
file.close()

from sklearn.feature_extraction.text import CountVectorizer
vectorizer = CountVectorizer(stop_words=stop_words)
X_train = vectorizer.fit_transform(raw_train_binary["分词文章"])
X_test = vectorizer.transform(raw_test_binary["分词文章"])

random_state=111
from sklearn.linear_model import SGDClassifier

percep_clf = SGDClassifier(loss="perceptron",penalty=None,learning_rate="constant",eta0=1.0,max_iter=1000,random_state=111)
lr_clf = SGDClassifier(loss="log",penalty=None,learning_rate="constant",eta0=1.0,max_iter=1000,random_state=111)
lsvm_clf = SGDClassifier(loss="hinge",penalty="l2",alpha=0.0001,learning_rate="constant",eta0=1.0,max_iter=1000,random_state=111)

# 训练感知机模型
percep_clf.fit(X_train,raw_train_binary["分类"])
# 输出测试集分类正确率
print(round(percep_clf.score(X_test,raw_test_binary["分类"]),2))

# 训练逻辑回归模型
lr_clf.fit(X_train,raw_train_binary["分类"])
# 输出测试集分类正确率
print(round(lr_clf.score(X_test,raw_test_binary["分类"]),2))

# 训练线性支持向量机模型
lsvm_clf.fit(X_train,raw_train_binary["分类"])
# 输出测试集分类正确率
print(round(lsvm_clf.score(X_test,raw_test_binary["分类"]),2))

from sklearn.metrics import confusion_matrix
import seaborn as sns
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(5,5))
# 设置正常显示中文
plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签
plt.rcParams['axes.unicode_minus']=False   #这两行需要手动设置
# 绘制热力图
y_svm_pred = lsvm_clf.predict(X_test) # 预测标签
y_test_true = raw_test_binary["分类"] #真实标签
confusion_matrix = confusion_matrix(y_svm_pred,y_test_true)#计算混淆矩阵
ax = sns.heatmap(confusion_matrix,linewidths=.5,cmap="Greens",
                 annot=True, fmt='d',xticklabels=lsvm_clf.classes_, yticklabels=lsvm_clf.classes_)
ax.set_ylabel('真实')
ax.set_xlabel('预测')
ax.xaxis.set_label_position('top')
ax.xaxis.tick_top()
ax.set_title('混淆矩阵热力图')
plt.show()