【SVM支持向量机】实现和案例

零、参考文章和资料

参考博客：这里
数据集：资料，提取码：zmlc
大佬笔记：资料，提取码：78ae

一、scitit-learn SVM API说明

1.1 算法库概述分类算法

svm.SVC API说明
参数说明：

1. C: 误差项的惩罚系数，默认为1.0；一般为大于0的一个数字，C越大表示在训练过程中对于总误差的关注度越高，也就是说当C越大的时候，对于训练集的表现会越好，但是有可能引发过度拟合的问题(overfiting)
2. kernel：指定SVM内部函数的类型，可选值：linear、poly、rbf、sigmoid、
3. precomputed(基本不用，有前提要求，要求特征属性数目和样本数目一样)；默认是rbf；
4. degree：当使用多项式函数作为svm内部的函数的时候，给定多项式的项数，默认为3
5. gamma：当SVM内部使用poly、rbf、sigmoid的时候，核函数的系数值，当默认值为auto的时候，实际系数为1/n_features
6. coef0: 当核函数为poly或者sigmoid的时候，给定的独立系数，默认为
7. probability：是否启用概率估计，默认不启动，不太建议启动
8. shrinking：是否开启收缩启发式计算，默认为True
9. tol: 模型构建收敛参数，当模型的的误差变化率小于该值的时候，结束模型构建过程，默认值:1e-3
10. cache_size：在模型构建过程中，缓存数据的最大内存大小，默认为空，单位MB
11. class_weight：给定各个类别的权重，默认为空
12. max_iter：最大迭代次数，默认-1表示不限制
13. decision_function_shape: 决策函数，可选值：ovo和ovr，默认为None；推荐使用ovr；（1.7以上版本才有）

1.2 scitit-learn SVM算法库概述回归算法

1.3 scitit-learn SVM-OneClassSVM

二、鸢尾花特征分类

# -*- coding: utf-8 -*

import numpy as np
import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
import warnings

from sklearn import svm #svm分类器
from sklearn.model_selection import train_test_split # 划分训练集 测试集
from sklearn.metrics import accuracy_score # 分类准确率
from sklearn.exceptions import ChangedBehaviorWarning

## 设置属性防止中文乱码
mpl.rcParams['font.sans-serif'] = [u'SimHei']
mpl.rcParams['axes.unicode_minus'] = False

warnings.filterwarnings('ignore', category=ChangedBehaviorWarning)

## 读取数据
# 'sepal length', 'sepal width', 'petal length', 'petal width'
iris_feature = u'花萼长度', u'花萼宽度', u'花瓣长度', u'花瓣宽度'
path = 'iris.csv'  # 数据文件路径
data = pd.read_csv(path, index_col=0,header=0,names=[0,1,2,3,4])

# 将类别装换成数字
mapping = {'setosa':0,'versicolor':1,'virginica':2}
data = data.replace(mapping)

x, y = data[list(range(4))], data[4]
x = x[[0, 1]] # 获取第一列和第二列


## 数据分割
x_train, x_test, y_train, y_test = train_test_split(x, y, random_state=0, train_size=0.8)

## 数据SVM分类器构建
clf = svm.SVC(C=1,kernel='rbf',gamma=0.1)
#gamma值越大，训练集的拟合就越好，但是会造成过拟合，导致测试集拟合变差
#gamma值越小，模型的泛化能力越好，训练集和测试集的拟合相近，但是会导致训练集出现欠拟合问题，从而，准确率变低，导致测试集准确率也变低。
#clf = svm.SVC(C=1, cache_size=200, class_weight=None, coef0=0.0,decision_function_shape=None, degree=3, gamma=0.1, kernel='rbf',
#  max_iter=-1, probability=False, random_state=None, shrinking=True, tol=0.001, verbose=False)
## 模型训练
clf.fit(x_train, y_train)

## 计算模型的准确率/精度
print (clf.score(x_train, y_train))
print ('训练集准确率：', accuracy_score(y_train, clf.predict(x_train)))
print (clf.score(x_test, y_test))
print ('测试集准确率：', accuracy_score(y_test, clf.predict(x_test)))

# 画图
N = 500
x1_min, x2_min = x.min()
x1_max, x2_max = x.max()
# print(x.max())
t1 = np.linspace(x1_min, x1_max, N)
t2 = np.linspace(x2_min, x2_max, N)
x1, x2 = np.meshgrid(t1, t2)  # 生成网格采样点
grid_show = np.dstack((x1.flat, x2.flat))[0] # 测试点

grid_hat = clf.predict(grid_show)       # 预测分类值
grid_hat = grid_hat.reshape(x1.shape)  # 使之与输入的形状相同

cm_light = mpl.colors.ListedColormap(['#00FFCC', '#FFA0A0', '#A0A0FF'])
cm_dark = mpl.colors.ListedColormap(['g', 'r', 'b'])
plt.figure(facecolor='w')
## 区域图
plt.pcolormesh(x1, x2, grid_hat, cmap=cm_light)
## 所有样本点
plt.scatter(x[0], x[1], c=y, edgecolors='k', s=50, cmap=cm_dark)      # 样本
## 测试数据集
plt.scatter(x_test[0], x_test[1], s=120, facecolors='none', zorder=10)     # 圈中测试集样本
## lable列表
plt.xlabel(iris_feature[0], fontsize=13)
plt.ylabel(iris_feature[1], fontsize=13)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'鸢尾花SVM特征分类', fontsize=16)
plt.grid(b=True, ls=':')
plt.tight_layout(pad=1.5)
plt.show()

三、鸢尾花数据集不同核函数效果比较

# -*- coding: utf-8 -*
import time
import numpy as np
import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
import warnings

from sklearn import svm  # svm导入
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.exceptions import ChangedBehaviorWarning

# 设置属性防止中文乱码
mpl.rcParams['font.sans-serif'] = [u'SimHei']
mpl.rcParams['axes.unicode_minus'] = False

warnings.filterwarnings('ignore', category=ChangedBehaviorWarning)

## 读取数据
# 'sepal length', 'sepal width', 'petal length', 'petal width'
iris_feature = u'花萼长度', u'花萼宽度', u'花瓣长度', u'花瓣宽度'
path = 'iris.csv'  # 数据文件路径
data = pd.read_csv(path, index_col=0,header=0,names=[0,1,2,3,4])

# 将类别装换成数字
mapping = {'setosa':0,'versicolor':1,'virginica':2}
data = data.replace(mapping)

x, y = data[list(range(4))], data[4]
x = x[[0, 1]] # 获取第一列和第二列

# 数据分割
x_train, x_test, y_train, y_test = train_test_split(
    x, y, random_state=0, train_size=0.8)

# 数据SVM分类器构建
svm1 = svm.SVC(C=1, kernel='linear')
svm2 = svm.SVC(C=1, kernel='rbf')
svm3 = svm.SVC(C=1, kernel='poly')
svm4 = svm.SVC(C=1, kernel='sigmoid')

# 模型训练
t0 = time.time()
svm1.fit(x_train, y_train)
t1 = time.time()
svm2.fit(x_train, y_train)
t2 = time.time()
svm3.fit(x_train, y_train)
t3 = time.time()
svm4.fit(x_train, y_train)
t4 = time.time()


# 效果评估
svm1_score1 = accuracy_score(y_train, svm1.predict(x_train))
svm1_score2 = accuracy_score(y_test, svm1.predict(x_test))

svm2_score1 = accuracy_score(y_train, svm2.predict(x_train))
svm2_score2 = accuracy_score(y_test, svm2.predict(x_test))

svm3_score1 = accuracy_score(y_train, svm3.predict(x_train))
svm3_score2 = accuracy_score(y_test, svm3.predict(x_test))

svm4_score1 = accuracy_score(y_train, svm4.predict(x_train))
svm4_score2 = accuracy_score(y_test, svm4.predict(x_test))


# 画图
x_tmp = [0, 1, 2, 3]
t_score = [t1 - t0, t2-t1, t3-t2, t4-t3]
y_score1 = [svm1_score1, svm2_score1, svm3_score1, svm4_score1]
y_score2 = [svm1_score2, svm2_score2, svm3_score2, svm4_score2]
plt.figure(facecolor='w', figsize=(12, 6))
plt.subplot(121)
plt.plot(x_tmp, y_score1, 'r-', lw=2, label=u'训练集准确率')
plt.plot(x_tmp, y_score2, 'g-', lw=2, label=u'测试集准确率')
plt.xlim(-0.3, 3.3)
plt.ylim(np.min((np.min(y_score1), np.min(y_score2)))*0.9,
          np.max((np.max(y_score1), np.max(y_score2)))*1.1)
plt.legend(loc='lower left')
plt.title(u'模型预测准确率', fontsize=13)
plt.xticks(x_tmp, [u'linear-SVM', u'rbf-SVM',
              u'poly-SVM', u'sigmoid-SVM'], rotation=0)
plt.grid(b=True)

plt.subplot(122)
plt.plot(x_tmp, t_score, 'b-', lw=2, label=u'模型训练时间')
plt.title(u'模型训练耗时', fontsize=13)
plt.xticks(x_tmp, [u'linear-SVM', u'rbf-SVM', u'poly-SVM', u'sigmoid-SVM'], rotation=0)
plt.xlim(-0.3, 3.3)
plt.grid(b=True)
plt.suptitle(u'鸢尾花数据SVM分类器不同内核函数模型比较', fontsize=16)

plt.show()


  # 预测结果画图
  # 画图比较
N = 500
x1_min, x2_min = x.min()
x1_max, x2_max = x.max()

t1 = np.linspace(x1_min, x1_max, N)
t2 = np.linspace(x2_min, x2_max, N)
x1, x2 = np.meshgrid(t1, t2)  # 生成网格采样点
grid_show = np.dstack((x1.flat, x2.flat))[0]  # 测试点

# 获取各个不同算法的测试值
svm1_grid_hat = svm1.predict(grid_show)
svm1_grid_hat = svm1_grid_hat.reshape(x1.shape)  # 使之与输入的形状相同

svm2_grid_hat = svm2.predict(grid_show)
svm2_grid_hat = svm2_grid_hat.reshape(x1.shape)  # 使之与输入的形状相同

svm3_grid_hat = svm3.predict(grid_show)
svm3_grid_hat = svm3_grid_hat.reshape(x1.shape)  # 使之与输入的形状相同

svm4_grid_hat = svm4.predict(grid_show)
svm4_grid_hat = svm4_grid_hat.reshape(x1.shape)  # 使之与输入的形状相同
 # 画图
cm_light = mpl.colors.ListedColormap(['#A0FFA0', '#FFA0A0', '#A0A0FF'])
cm_dark = mpl.colors.ListedColormap(['g', 'r', 'b'])
plt.figure(facecolor='w', figsize=(14, 7))

# svm
plt.subplot(221)
# 区域图
# plt.pcolormesh(x1, x2, svm1_grid_hat, cmap=cm_light)
# 所以样本点
# 样本
plt.scatter(x[0], x[1], c=y, edgecolors='k', s=50, cmap=cm_dark)
# 测试数据集
# 圈中测试集样本
plt.scatter(x_test[0], x_test[1], s=120, facecolors='none', zorder=10)
# lable列表
plt.xlabel(iris_feature[0], fontsize=13)
plt.ylabel(iris_feature[1], fontsize=13)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'鸢尾花Linear-SVM特征分类', fontsize=16)
plt.grid(b=True, ls=':')
plt.tight_layout(pad=1.5)

plt.subplot(222)
# 区域图
plt.pcolormesh(x1, x2, svm2_grid_hat, cmap=cm_light)
# 所以样本点
# 样本
plt.scatter(x[0], x[1], c=y, edgecolors='k', s=50, cmap=cm_dark)
# 测试数据集
# 圈中测试集样本
plt.scatter(x_test[0], x_test[1], s=120, facecolors='none', zorder=10)
# lable列表
plt.xlabel(iris_feature[0], fontsize=13)
plt.ylabel(iris_feature[1], fontsize=13)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'鸢尾花rbf-SVM特征分类', fontsize=16)
plt.grid(b=True, ls=':')
plt.tight_layout(pad=1.5)

plt.subplot(223)
 # 区域图
plt.pcolormesh(x1, x2, svm3_grid_hat, cmap=cm_light)
# 所以样本点
# 样本
plt.scatter(x[0], x[1], c=y, edgecolors='k', s=50, cmap=cm_dark)
# 测试数据集
# 圈中测试集样本
plt.scatter(x_test[0], x_test[1], s=120, facecolors='none', zorder=10)
# lable列表
plt.xlabel(iris_feature[0], fontsize=13)
plt.ylabel(iris_feature[1], fontsize=13)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'鸢尾花poly-SVM特征分类', fontsize=16)
plt.grid(b=True, ls=':')
plt.tight_layout(pad=1.5)

plt.subplot(224)
# 区域图
plt.pcolormesh(x1, x2, svm4_grid_hat, cmap=cm_light)
# 所以样本点
# 样本
plt.scatter(x[0], x[1], c=y, edgecolors='k', s=50, cmap=cm_dark)
# 测试数据集
# 圈中测试集样本
plt.scatter(x_test[0], x_test[1], s=120, facecolors='none', zorder=10)
# lable列表
plt.xlabel(iris_feature[0], fontsize=13)
plt.ylabel(iris_feature[1], fontsize=13)
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title(u'鸢尾花sigmoid-SVM特征分类', fontsize=16)
plt.grid(b=True, ls=':')
plt.tight_layout(pad=1.5)
plt.show()

四、BIM分类

数据产生

import random

def calc_bmi(h, w):
    bmi = w / (h/100) ** 2
    if bmi < 18.5: return "thin"
    if bmi < 25.0: return "normal"
    return "fat"

fp = open("./data/bmi.csv","w",encoding="utf-8")
fp.write("height,weight,label\r\n")

cnt = {"thin":0, "normal":0, "fat":0}
for i in range(20000):
    h = random.randint(120,200)
    w = random.randint(35, 80)
    label = calc_bmi(h, w)
    cnt[label] += 1
    fp.write("{0},{1},{2}\r\n".format(h, w, label))
fp.close()
print("ok,", cnt)

原始数据的可视化

import matplotlib.pyplot as plt
import pandas as pd

tbl = pd.read_csv("./data/bmi.csv", index_col=2)

fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)

def scatter(lbl, color):
    b = tbl.loc[lbl]
    ax.scatter(b["weight"],b["height"], c=color, label=lbl)

scatter("fat",    "red")
scatter("normal", "yellow")
scatter("thin",   "purple")

ax.legend() 
plt.show()

svm分类

from sklearn.svm import SVC #svm分类器
from sklearn.model_selection import train_test_split # 划分训练集 测试集
from sklearn.metrics import accuracy_score # 分类准确率
import matplotlib.pyplot as plt
import pandas as pd

tbl = pd.read_csv("./data/bmi.csv")

label = tbl["label"]
w = tbl["weight"] / 100 
h = tbl["height"] / 200 
wh = pd.concat([w, h], axis=1)

data_train, data_test, label_train, label_test = train_test_split(wh, label)

clf = SVC()
clf.fit(data_train, label_train)

predict = clf.predict(data_test)

ac_score = accuracy_score(label_test, predict)
print(ac_score)

分类结果可视化