文本分类实验
http://qwone.com/~jason/20Newsgroups/下载数据
鸢尾花
Iris_GaussianNB.py
#!/usr/bin/python
# -*- coding:utf-8 -*-
import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl
from sklearn.preprocessing import StandardScaler
from sklearn.naive_bayes import GaussianNB, MultinomialNB
from sklearn.pipeline import Pipeline
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import train_test_split # 分数据集
def iris_type(s):
it = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}
return it[s]
if __name__ == "__main__":
data = np.loadtxt('..\8.Regression\8.iris.data', dtype=float, delimiter=',', converters={4: iris_type})
print data
x, y = np.split(data, (4,), axis=1)
x = x[:, :2]
print x
print y
# train_test_split(x,y,train_size=0.8)
gnb = Pipeline([
('sc', StandardScaler()), # 标准化
('clf', GaussianNB())]) #假定服从高斯分布,朴素贝叶斯
gnb.fit(x, y.ravel()) # 转成向量
# gnb = MultinomialNB().fit(x, y.ravel())
# gnb = KNeighborsClassifier(n_neighbors=5).fit(x, y.ravel()) # K近邻,设置层数
# 画图
N, M = 500, 500 # 横纵各采样多少个值
x1_min, x1_max = x[:, 0].min(), x[:, 0].max() # 第0列的范围
x2_min, x2_max = x[:, 1].min(), x[:, 1].max() # 第1列的范围
t1 = np.linspace(x1_min, x1_max, N)
t2 = np.linspace(x2_min, x2_max, M)
x1, x2 = np.meshgrid(t1, t2) # 生成网格采样点
x_test = np.stack((x1.flat, x2.flat), axis=1) # 测试点
# 无意义,只是为了凑另外两个维度
# x3 = np.ones(x1.size) * np.average(x[:, 2])
# x4 = np.ones(x1.size) * np.average(x[:, 3])
# x_test = np.stack((x1.flat, x2.flat, x3, x4), axis=1) # 测试点
mpl.rcParams['font.sans-serif'] = [u'simHei']
mpl.rcParams['axes.unicode_minus'] = False
cm_light = mpl.colors.ListedColormap(['#77E0A0', '#FF8080', '#A0A0FF'])
cm_dark = mpl.colors.ListedColormap(['g', 'r', 'b'])
y_hat = gnb.predict(x_test) # 预测值
y_hat = y_hat.reshape(x1.shape) # 使之与输入的形状相同
plt.figure(facecolor='w')
plt.pcolormesh(x1, x2, y_hat, cmap=cm_light) # 预测值的显示
plt.scatter(x[:, 0], x[:, 1], c=y, edgecolors='k', s=50, cmap=cm_dark) # 样本的显示
plt.xlabel(u'花萼长度', fontsize=14)
plt.ylabel(u'花萼宽度', fontsize=14)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'GaussianNB对鸢尾花数据的分类结果', fontsize=18)
plt.grid(True)
plt.show()
# 训练集上的预测结果
y_hat = gnb.predict(x)
y = y.reshape(-1)
result = y_hat == y
print y_hat
print result
acc = np.mean(result)
print '准确度: %.2f%%' % (100 * acc)
MultinomialNB_intro.py
#!/usr/bin/python
# -*- coding:utf-8 -*-
import numpy as np
from sklearn.naive_bayes import GaussianNB, MultinomialNB
if __name__ == "__main__":
np.random.seed(0) # 随机数据一样
M = 20 # 20样本
N = 5 # 五维
x = np.random.randint(2, size=(M, N)) # 只能取0和1的MN矩阵,[low, high)
x = np.array(list(set([tuple(t) for t in x]))) # 把数据提出变成元组,放入集合,再转成列表,形成新的数据----去重,20行除去重复的成为17行
M = len(x) # 提取行数
y = np.arange(M) # 行数作为y的标记值,每一行就是一个类别
print '样本个数:%d,特征数目:%d' % x.shape
print '样本:\n', x
mnb = MultinomialNB(alpha=1) # 朴素贝叶斯;动手:换成GaussianNB()试试预测结果?
# mnb = GaussianNB()
mnb.fit(x, y) # 训练
y_hat = mnb.predict(x) # 预测,看y和y_hat是否符合
print '预测类别:', y_hat
print '准确率:%.2f%%' % (100*np.mean(y_hat == y))
print '系统得分:', mnb.score(x, y)
# from sklearn import metrics
# print metrics.accuracy_score(y, y_hat)
err = y_hat != y
for i, e in enumerate(err): # 枚举错误的值,y是实际的,y_hat是被认为的
if e:
print y[i], ':\t', x[i], '被认为与', x[y_hat[i]], '一个类别'
六模型文本处理
text_classification.py
#!/usr/bin/python
# -*- coding:utf-8 -*-
import numpy as np
from sklearn.naive_bayes import MultinomialNB, BernoulliNB
from sklearn.datasets import fetch_20newsgroups # 函数中包含了数据
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import RidgeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV
from sklearn import metrics
from time import time
from pprint import pprint
import matplotlib.pyplot as plt
import matplotlib as mpl
def test_clf(clf): # 某个分类器
print u'分类器:', clf
alpha_can = np.logspace(-3, 2, 10) # 等比数列,取10个不同的参数值
model = GridSearchCV(clf, param_grid={'alpha': alpha_can}, cv=5) # 一个分类器做五次交叉验证(网格搜索),共50次
m = alpha_can.size # 用于计算时间
if hasattr(clf, 'alpha'): # NB的超参数
model.set_params(param_grid={'alpha': alpha_can})
m = alpha_can.size
if hasattr(clf, 'n_neighbors'): # k-近邻的超参数
neighbors_can = np.arange(1, 15)
model.set_params(param_grid={'n_neighbors': neighbors_can})
m = neighbors_can.size
if hasattr(clf, 'C'): # SVM超参数
C_can = np.logspace(1, 3, 3) # 时间长不多取
gamma_can = np.logspace(-3, 0, 3)
model.set_params(param_grid={'C':C_can, 'gamma':gamma_can})
m = C_can.size * gamma_can.size
if hasattr(clf, 'max_depth'): # 随机森林超参数
max_depth_can = np.arange(4, 10)
model.set_params(param_grid={'max_depth': max_depth_can})
m = max_depth_can.size
t_start = time()
model.fit(x_train, y_train) # 训练,得到最佳参数
t_end = time()
t_train = (t_end - t_start) / (5*m) # 除以词次数
print u'5折交叉验证的训练时间为:%.3f秒/(5*%d)=%.3f秒' % ((t_end - t_start), m, t_train)
print u'最优超参数为:', model.best_params_ # 打印最佳超参数
t_start = time()
y_hat = model.predict(x_test)
t_end = time()
t_test = t_end - t_start
print u'测试时间:%.3f秒' % t_test
acc = metrics.accuracy_score(y_test, y_hat)
print u'测试集准确率:%.2f%%' % (100 * acc)
name = str(clf).split('(')[0] # 转成字符串
index = name.find('Classifier')
if index != -1: # 如果没找到
name = name[:index] # 去掉末尾的Classifier
if name == 'SVC': # SVC换成SVM
name = 'SVM'
return t_train, t_test, 1-acc, name # 返回这个分类器的最终结果
if __name__ == "__main__":
print u'开始下载/加载数据...'
t_start = time()
# remove = ('headers', 'footers', 'quotes') # 删除标题等等
remove = ()
categories = 'alt.atheism', 'talk.religion.misc', 'comp.graphics', 'sci.space' # 关注的四个类别
# categories = None # 若分类所有类别,请注意内存是否够用
data_train = fetch_20newsgroups(subset='train', categories=categories, shuffle=True, random_state=0, remove=remove)
data_test = fetch_20newsgroups(subset='test', categories=categories, shuffle=True, random_state=0, remove=remove)
t_end = time() # 显示下载时间
print u'下载/加载数据完成,耗时%.3f秒' % (t_end - t_start)
print u'数据类型:', type(data_train)
print u'训练集包含的文本数目:', len(data_train.data)
print u'测试集包含的文本数目:', len(data_test.data)
print u'训练集和测试集使用的%d个类别的名称:' % len(categories) # 4
categories = data_train.target_names
pprint(categories) # 效果稍好
y_train = data_train.target # 类别
y_test = data_test.target
# 查看数据
print u' -- 前10个文本 -- '
for i in np.arange(10):
print u'文本%d(属于类别 - %s):' % (i+1, categories[y_train[i]]) # 显示类别
print data_train.data[i] # 训练数据文本
print '\n\n'
# 提取词频做特征,非必须
vectorizer = TfidfVectorizer(input='content', stop_words='english', max_df=0.5, sublinear_tf=True) # TF-TDF模型提取特征,设置停止词,出现频率
x_train = vectorizer.fit_transform(data_train.data) # 做特征转换;x_train是稀疏的,scipy.sparse.csr.csr_matrix
x_test = vectorizer.transform(data_test.data) # 测试数据特征转换
print u'训练集样本个数:%d,特征个数:%d' % x_train.shape # 样本个数即文本数,特征个数即词数
print u'停止词:\n',
pprint(vectorizer.get_stop_words()) # 停止词(非重要)
feature_names = np.asarray(vectorizer.get_feature_names())
print u'\n\n===================\n分类器的比较:\n'
# 运行时间,一次时间,正确率
clfs = (MultinomialNB(), # 0.87(0.017), 0.002, 90.39%
BernoulliNB(), # 1.592(0.032), 0.010, 88.54%
KNeighborsClassifier(), # 19.737(0.282), 0.208, 86.03%
RidgeClassifier(), # 25.6(0.512), 0.003, 89.73%
RandomForestClassifier(n_estimators=200), # 59.319(1.977), 0.248, 77.01%
SVC() # 236.59(5.258), 1.574, 90.10%,允许错误的SVM
)
result = []
for clf in clfs: # 做遍历
a = test_clf(clf)
result.append(a) # 放到result中
print '\n'
result = np.array(result)
time_train, time_test, err, names = result.T
x = np.arange(len(time_train))
mpl.rcParams['font.sans-serif'] = [u'simHei']
mpl.rcParams['axes.unicode_minus'] = False
plt.figure(figsize=(10, 7), facecolor='w')
ax = plt.axes()
b1 = ax.bar(x, err, width=0.25, color='#77E0A0')
ax_t = ax.twinx()
b2 = ax_t.bar(x+0.25, time_train, width=0.25, color='#FFA0A0')
b3 = ax_t.bar(x+0.5, time_test, width=0.25, color='#FF8080')
plt.xticks(x+0.5, names, fontsize=10)
leg = plt.legend([b1[0], b2[0], b3[0]], (u'错误率', u'训练时间', u'测试时间'), loc='upper left', shadow=True)
# for lt in leg.get_texts():
# lt.set_fontsize(14)
plt.title(u'新闻组文本数据不同分类器间的比较', fontsize=18)
plt.xlabel(u'分类器名称')
plt.grid(True)
plt.tight_layout(2)
plt.show()