1. SVM多分类案例
SVM从原理而言只能做二分类任务。但是如果我们同时使用多个SVM去对数据进行拟合,就可以实现多分类的任务了。在sklearn包中,我们可以选择两种方式( decision_function_shape参数)去实现多分类任务,第一种是“ovo”,即一对一模式,如我们要将数据分为四类,便需要建立六个分类器(1vs2,1vs3,1vs4,2vs3,2vs4,3vs4)。第二种是“ovr”模型,即一对其他模型,如我们要将数据分为四类,便需要建立四个分类器(1 vs others,2 vs others,3 vs others,4 vs others)。这下面的案例中,我们生成了三组多元正态随机数,来进行三分类训练。
import numpy as np
from sklearn import svm
from scipy import stats
from sklearn.metrics import accuracy_score
import matplotlib as mpl
import matplotlib.pyplot as plt
def extend(a, b, r):
x = a - b
m = (a + b) / 2
return m-r*x/2, m+r*x/2
if __name__ == "__main__":
np.random.seed(0)
N = 20
x = np.empty((4*N, 2))
means = [(-1, 1), (1, 1), (1, -1), (-1, -1)]
sigmas = [np.eye(2), 2*np.eye(2), np.diag((1,2)), np.array(((2,1),(1,2)))]
for i in range(4):
mn = stats.multivariate_normal(means[i], sigmas[i]*0.3)
x[i*N:(i+1)*N, :] = mn.rvs(N)
a = np.array((0,1,2,3)).reshape((-1, 1))
y = np.tile(a, N).flatten()
clf = svm.SVC(C=1, kernel='rbf', gamma=1, decision_function_shape='ovo')
# clf = svm.SVC(C=1, kernel='linear', decision_function_shape='ovr')
clf.fit(x, y)
y_hat = clf.predict(x)
acc = accuracy_score(y, y_hat)
np.set_printoptions(suppress=True)
print (u'预测正确的样本个数:%d,正确率:%.2f%%' % (round(acc*4*N), 100*acc))
# decision_function
print (clf.decision_function(x))
print (y_hat)
x1_min, x2_min = np.min(x, axis=0)
x1_max, x2_max = np.max(x, axis=0)
x1_min, x1_max = extend(x1_min, x1_max, 1.05)
x2_min, x2_max = extend(x2_min, x2_max, 1.05)
x1, x2 = np.mgrid[x1_min:x1_max:500j, x2_min:x2_max:500j]
x_test = np.stack((x1.flat, x2.flat), axis=1)
y_test = clf.predict(x_test)
y_test = y_test.reshape(x1.shape)
cm_light = mpl.colors.ListedColormap(['#FF8080', '#A0FFA0', '#6060FF', '#F080F0'])
cm_dark = mpl.colors.ListedColormap(['r', 'g', 'b', 'm'])
mpl.rcParams['font.sans-serif'] = [u'SimHei']
mpl.rcParams['axes.unicode_minus'] = False
plt.figure(facecolor='w')
plt.pcolormesh(x1, x2, y_test, cmap=cm_light)
plt.scatter(x[:, 0], x[:, 1], s=40, c=y, cmap=cm_dark, alpha=0.7)
plt.xlim((x1_min, x1_max))
plt.ylim((x2_min, x2_max))
plt.grid(b=True)
plt.tight_layout(pad=2.5)
plt.title(u'SVM多分类方法:One/One or One/Other', fontsize=18)
plt.show()
最终结果如下:
2. SVM 参数的解读
SVM模型的参数较少。如果选择线性核,那主要需要调的参数便是C,也即惩罚因子,从图中第一行可以看出,C越大分离超平面的两条虚线越接近。而对于高斯核来说,需要调的参数主要有两个,一个是C,一个是
γ
\gamma
γ,其中
γ
\gamma
γ是高斯核函数中的系数,与方差有关。从图中可以看出,
γ
\gamma
γ越大,SVM对于训练数据的拟合越充分,但是同时可能会带来过拟合的问题。
代码如下:
import numpy as np
from sklearn import svm
import matplotlib as mpl
import matplotlib.colors
import matplotlib.pyplot as plt
def show_accuracy(a, b):
acc = a.ravel() == b.ravel()
if __name__ == "__main__":
data = np.loadtxt('bipartition.txt', dtype=np.float, delimiter='\t')
x, y = np.split(data, (2, ), axis=1)
y = y.ravel()
# 分类器
clf_param = (('linear', 0.1), ('linear', 0.5), ('linear', 1), ('linear', 2),
('rbf', 1, 0.1), ('rbf', 1, 1), ('rbf', 1, 10), ('rbf', 1, 100),
('rbf', 5, 0.1), ('rbf', 5, 1), ('rbf', 5, 10), ('rbf', 5, 100))
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
x1, x2 = np.mgrid[x1_min:x1_max:200j, x2_min:x2_max:200j] # 生成网格采样点
grid_test = np.stack((x1.flat, x2.flat), axis=1) # 测试点
cm_light = mpl.colors.ListedColormap(['#77E0A0', '#FFA0A0'])
cm_dark = mpl.colors.ListedColormap(['g', 'r'])
mpl.rcParams['font.sans-serif'] = [u'SimHei']
mpl.rcParams['axes.unicode_minus'] = False
plt.figure(figsize=(14, 10), facecolor='w')
for i, param in enumerate(clf_param):
clf = svm.SVC(C=param[1], kernel=param[0])
if param[0] == 'rbf':
clf.gamma = param[2]
title = u'高斯核,C=%.1f,$\gamma$ =%.1f' % (param[1], param[2])
else:
title = u'线性核,C=%.1f' % param[1]
clf.fit(x, y)
y_hat = clf.predict(x)
show_accuracy(y_hat, y) # 准确率
# 画图
print (title)
print ('支撑向量的数目:', clf.n_support_)
print ('支撑向量的系数:', clf.dual_coef_)
print ('支撑向量:', clf.support_)
plt.subplot(3, 4, i+1)
grid_hat = clf.predict(grid_test) # 预测分类值
grid_hat = grid_hat.reshape(x1.shape) # 使之与输入的形状相同
plt.pcolormesh(x1, x2, grid_hat, cmap=cm_light, alpha=0.8)
plt.scatter(x[:, 0], x[:, 1], c=y, edgecolors='k', s=40, cmap=cm_dark) # 样本的显示
plt.scatter(x[clf.support_, 0], x[clf.support_, 1], edgecolors='k', facecolors='none', s=100, marker='o') # 支撑向量
z = clf.decision_function(grid_test)
print ('clf.decision_function(x) = ', clf.decision_function(x))
print ('clf.predict(x) = ', clf.predict(x))
z = z.reshape(x1.shape)
plt.contour(x1, x2, z, colors=list('kbrbk'), linestyles=['--', '--', '-', '--', '--'],
linewidths=[1, 0.5, 1.5, 0.5, 1], levels=[-1, -0.5, 0, 0.5, 1])
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(title, fontsize=14)
plt.suptitle(u'SVM不同参数的分类', fontsize=20)
plt.tight_layout(1.4)
plt.subplots_adjust(top=0.92)
plt.savefig('1.png')
plt.show()
3. SVM不平衡数据的处理
SVM可以通过调节参数进行不平衡数据的处理。在下例中,正样本只有10个,而负样本有990个,我们通过调整class_weight参数,如:class_weight={-1: 1, 1: 10},来拟合不平衡的数据。
代码如下:
import numpy as np
from sklearn import svm
import matplotlib.colors
import matplotlib.pyplot as plt
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.exceptions import UndefinedMetricWarning
import warnings
if __name__ == "__main__":
warnings.filterwarnings(action='ignore', category=UndefinedMetricWarning)
np.random.seed(0) # 保持每次生成的数据相同
c1 = 990
c2 = 10
N = c1 + c2
x_c1 = 3*np.random.randn(c1, 2)
x_c2 = 0.5*np.random.randn(c2, 2) + (4, 4)
x = np.vstack((x_c1, x_c2))
y = np.ones(N)
y[:c1] = -1
# 显示大小
s = np.ones(N) * 30
s[:c1] = 10
# 分类器
clfs = [svm.SVC(C=1, kernel='linear'),
svm.SVC(C=1, kernel='linear', class_weight={-1: 1, 1: 50}),
svm.SVC(C=0.8, kernel='rbf', gamma=0.5, class_weight={-1: 1, 1: 2}),
svm.SVC(C=0.8, kernel='rbf', gamma=0.5, class_weight={-1: 1, 1: 10})]
titles = 'Linear', 'Linear, Weight=50', 'RBF, Weight=2', 'RBF, Weight=10'
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
x1, x2 = np.mgrid[x1_min:x1_max:500j, x2_min:x2_max:500j] # 生成网格采样点
grid_test = np.stack((x1.flat, x2.flat), axis=1) # 测试点
cm_light = matplotlib.colors.ListedColormap(['#77E0A0', '#FF8080'])
cm_dark = matplotlib.colors.ListedColormap(['g', 'r'])
matplotlib.rcParams['font.sans-serif'] = [u'SimHei']
matplotlib.rcParams['axes.unicode_minus'] = False
plt.figure(figsize=(10, 8), facecolor='w')
for i, clf in enumerate(clfs):
clf.fit(x, y)
y_hat = clf.predict(x)
# show_accuracy(y_hat, y) # 正确率
# show_recall(y, y_hat) # 召回率
print (i+1, '次:')
print ('accuracy:\t', accuracy_score(y, y_hat))
print ('precision:\t', precision_score(y, y_hat, pos_label=1))
print ('recall:\t', recall_score(y, y_hat, pos_label=1))
print ('F1-score:\t', f1_score(y, y_hat, pos_label=1))
# 画图
plt.subplot(2, 2, i+1)
grid_hat = clf.predict(grid_test) # 预测分类值
grid_hat = grid_hat.reshape(x1.shape) # 使之与输入的形状相同
plt.pcolormesh(x1, x2, grid_hat, cmap=cm_light, alpha=0.8)
plt.scatter(x[:, 0], x[:, 1], c=y, edgecolors='k', s=s, cmap=cm_dark) # 样本的显示
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(titles[i])
plt.grid()
plt.suptitle(u'不平衡数据的处理', fontsize=18)
plt.tight_layout(1.5)
plt.subplots_adjust(top=0.92)
plt.show()
4. MNIST手写数据识别
在这一部分,我们通过手写数据来对SVM的分类效果进行测试,对照组选取的是随机森林。MNIST手写数据也是很著名的分类数据。具体形式如下图:
代码如下:
import numpy as np
from sklearn import svm
import matplotlib.colors
import matplotlib.pyplot as plt
from PIL import Image
from sklearn.metrics import accuracy_score
import pandas as pd
import os
import csv
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from time import time
def save_image(im, i):
im = 255 - im
a = im.astype(np.uint8)
output_path = '.\\HandWritten'
if not os.path.exists(output_path):
os.mkdir(output_path)
Image.fromarray(a).save(output_path + ('\\%d.png' % i))
def save_result(model):
data_test_hat = model.predict(data_test)
with open('Prediction.csv', 'wb') as f:
writer = csv.writer(f)
writer.writerow(['ImageId', 'Label'])
for i, d in enumerate(data_test_hat):
writer.writerow([i, d])
# writer.writerows(zip(np.arange(1, len(data_test_hat) + 1), data_test_hat))
if __name__ == "__main__":
classifier_type = 'RF'
print ('载入训练数据...')
t = time()
data = pd.read_csv('.\\MNIST.train.csv', header=0, dtype=np.int)
print ('载入完成,耗时%f秒' % (time() - t))
y = data['label'].values
x = data.values[:, 1:]
print ('图片个数:%d,图片像素数目:%d' % x.shape)
images = x.reshape(-1, 28, 28)
y = y.ravel()
print ('载入测试数据...')
t = time()
data_test = pd.read_csv('.\\MNIST.test.csv', header=0, dtype=np.int)
data_test = data_test.values
images_test_result = data_test.reshape(-1, 28, 28)
print ('载入完成,耗时%f秒' % (time() - t))
np.random.seed(0)
x, x_test, y, y_test = train_test_split(x, y, train_size=0.8, random_state=1)
images = x.reshape(-1, 28, 28)
images_test = x_test.reshape(-1, 28, 28)
print (x.shape, x_test.shape)
matplotlib.rcParams['font.sans-serif'] = [u'SimHei']
matplotlib.rcParams['axes.unicode_minus'] = False
plt.figure(figsize=(15, 9), facecolor='w')
for index, image in enumerate(images[:16]):
plt.subplot(4, 8, index + 1)
plt.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
plt.title(u'训练图片: %i' % y[index])
for index, image in enumerate(images_test_result[:16]):
plt.subplot(4, 8, index + 17)
plt.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
save_image(image.copy(), index)
plt.title(u'测试图片')
plt.tight_layout()
plt.show()
# SVM
if classifier_type == 'SVM':
# params = {'C':np.logspace(1, 4, 4, base=10), 'gamma':np.logspace(-10, -2, 9, base=10)}
# clf = svm.SVC(kernel='rbf')
# model = GridSearchCV(clf, param_grid=params, cv=3)
model = svm.SVC(C=1000, kernel='rbf', gamma=1e-10)
print ('SVM开始训练...')
t = time()
model.fit(x, y)
t = time() - t
print ('SVM训练结束,耗时%d分钟%.3f秒' % (int(t/60), t - 60*int(t/60)))
# print '最优分类器:', model.best_estimator_
# print '最优参数:\t', model.best_params_
# print 'model.cv_results_ ='
# pprint(model.cv_results_)
t = time()
y_hat = model.predict(x)
t = time() - t
print ('SVM训练集准确率:%.3f%%,耗时%d分钟%.3f秒' % (accuracy_score(y, y_hat)*100, int(t/60), t - 60*int(t/60)))
t = time()
y_test_hat = model.predict(x_test)
t = time() - t
print ('SVM测试集准确率:%.3f%%,耗时%d分钟%.3f秒' % (accuracy_score(y_test, y_test_hat)*100, int(t/60), t - 60*int(t/60)))
save_result(model)
elif classifier_type == 'RF':
rfc = RandomForestClassifier(100, criterion='gini', min_samples_split=2,
min_impurity_split=1e-10, bootstrap=True, oob_score=True)
print ('随机森林开始训练...')
t = time()
rfc.fit(x, y)
t = time() - t
print ('随机森林训练结束,耗时%d分钟%.3f秒' % (int(t/60), t - 60*int(t/60)))
print ('OOB准确率:%.3f%%' % (rfc.oob_score_*100))
t = time()
y_hat = rfc.predict(x)
t = time() - t
print ('随机森林训练集准确率:%.3f%%,预测耗时:%d秒' % (accuracy_score(y, y_hat)*100, t))
t = time()
y_test_hat = rfc.predict(x_test)
t = time() - t
print ('随机森林测试集准确率:%.3f%%,预测耗时:%d秒' % (accuracy_score(y_test, y_test_hat)*100, t))
save_result(rfc)
err = (y_test != y_test_hat)
err_images = images_test[err]
err_y_hat = y_test_hat[err]
err_y = y_test[err]
print (err_y_hat)
print (err_y)
plt.figure(figsize=(10, 8), facecolor='w')
for index, image in enumerate(err_images):
if index >= 12:
break
plt.subplot(3, 4, index + 1)
plt.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
plt.title(u'错分为:%i,真实值:%i' % (err_y_hat[index], err_y[index]))
plt.suptitle(u'数字图片手写体识别:分类器%s' % classifier_type, fontsize=18)
plt.tight_layout(rect=(0, 0, 1, 0.95))
plt.show()