题目要求：6.3 选择两个 UCI 数据集，分别用线性核和高斯核训练一个 SVM，并与BP 神经网络和 C4.5 决策树进行实验比较。

运行结果如下：

全部代码如下：

# 导入库
import numpy as np
import pandas as pd
from sklearn import datasets
from sklearn.model_selection import KFold, train_test_split, cross_val_score, cross_validate
from sklearn import svm, tree, model_selection, metrics, preprocessing
from c45 import C45

# 读入第一组数据
iris = datasets.load_iris()
X = pd.DataFrame(iris['data'], columns=iris['feature_names'])
y = pd.Series(iris['target_names'][iris['target']])
# 数据样例展示
print(X.head())

# 训练线性核SVM
linear_svm = svm.SVC(C=1, kernel='linear')
linear_scores = cross_validate(linear_svm, X, y, cv=5, scoring='accuracy')
print(linear_scores['test_score'].mean())  # 训练结果

# 训练高斯核SVM
rbf_svm = svm.SVC(C=1, kernel='rbf')
rbf_scores = cross_validate(rbf_svm, X, y, cv=5, scoring='accuracy')
print(rbf_scores['test_score'].mean())  # 训练结果

# C4.5算法
clf = C45(attrNames=iris.feature_names)
X_train, X_test, y_train, y_test = train_test_split(iris.data, iris.target, test_size=0.5)
clf.fit(X_train, y_train)
print(clf.score(X_test, y_test))

# 读入第二组数据
bre_c = datasets.load_breast_cancer()
X = pd.DataFrame(bre_c['data'], columns=bre_c['feature_names'])
y = pd.Series(bre_c['target_names'][bre_c['target']])
# 数据样例展示
print(X.head())

# 训练线性核SVM
linear_svm = svm.SVC(C=1, kernel='linear')
linear_scores = cross_validate(linear_svm, X, y, cv=5, scoring='accuracy')
print(linear_scores['test_score'].mean())  # 训练结果

# 训练高斯核SVM
rbf_svm = svm.SVC(C=1, kernel='rbf')
rbf_scores = cross_validate(rbf_svm, X, y, cv=5, scoring='accuracy')
print(rbf_scores['test_score'].mean())  # 训练结果

# C4.5算法
clf = C45(attrNames=bre_c.feature_names)
X_train, X_test, y_train, y_test = train_test_split(bre_c.data, bre_c.target, test_size=0.5)
clf.fit(X_train, y_train)
print(clf.score(X_test, y_test))

程序解释说明：

数据导入

使用sklearn自带的uci数据集进行测试，并打印展示

# 读入第一组数据
iris = datasets.load_iris()
X = pd.DataFrame(iris['data'], columns=iris['feature_names'])
y = pd.Series(iris['target_names'][iris['target']])
# 数据样例展示
print(X.head())

线性核和高斯核

使用sklearn的SVM进行训练，并打印训练结果

# 训练线性核SVM
linear_svm = svm.SVC(C=1, kernel='linear')
linear_scores = cross_validate(linear_svm, X, y, cv=5, scoring='accuracy')
print(linear_scores['test_score'].mean())  # 训练结果

# 训练高斯核SVM
rbf_svm = svm.SVC(C=1, kernel='rbf')
rbf_scores = cross_validate(rbf_svm, X, y, cv=5, scoring='accuracy')
print(rbf_scores['test_score'].mean())  # 训练结果

C4.5算法

从github上下载了C4.5算法的数据库

https://github.com/RaczeQ/scikit-learn-C4.5-tree-classifier

将数据库导入site-package文件夹后，可直接进行使用。

这里给出其中实现算法的代码

import math
from xml.dom import minidom
from xml.etree import ElementTree as ET

from sklearn.base import BaseEstimator, ClassifierMixin
from sklearn.utils.validation import check_array, check_is_fitted, check_X_y

from .c45_utils import decision, grow_tree

class C45(BaseEstimator, ClassifierMixin):
    """A C4.5 tree classifier.

    Parameters
    ----------
    attrNames : list, optional (default=None)
        The list of feature names used in printing tree during. If left default,
        attributes will be named attr0, attr1... etc
    See also
    --------
    DecisionTreeClassifier
    References
    ----------
    .. [1] https://en.wikipedia.org/wiki/Decision_tree_learning
    .. [2] https://en.wikipedia.org/wiki/C4.5_algorithm
    .. [3] L. Breiman, J. Friedman, R. Olshen, and C. Stone, "Classification
           and Regression Trees", Wadsworth, Belmont, CA, 1984.
    .. [4] J. R. Quinlain, "C4.5: Programs for Machine Learning",
           Morgan Kaufmann Publishers, 1993
    Examples
    --------
    >>> from sklearn.datasets import load_iris
    >>> from sklearn.model_selection import cross_val_score
    >>> from c45 import C45
    >>> iris = load_iris()
    >>> clf = C45(attrNames=iris.feature_names)
    >>> cross_val_score(clf, iris.data, iris.target, cv=10)
    ...                             # doctest: +SKIP
    ...
    array([ 1.     ,  0.93...,  0.86...,  0.93...,  0.93...,
            0.93...,  0.93...,  1.     ,  0.93...,  1.      ])
    """
    def __init__(self, attrNames=None):
        if attrNames is not None:
            attrNames = [''.join(i for i in x if i.isalnum()).replace(' ', '_') for x in attrNames]
        self.attrNames = attrNames

    def fit(self, X, y):
        X, y = check_X_y(X, y)
        self.X_ = X
        self.y_ = y
        self.resultType = type(y[0])
        if self.attrNames is None:
            self.attrNames = [f'attr{x}' for x in range(len(self.X_[0]))]

        assert(len(self.attrNames) == len(self.X_[0]))

        data = [[] for i in range(len(self.attrNames))]
        categories = []

        for i in range(len(self.X_)):
            categories.append(str(self.y_[i]))
            for j in range(len(self.attrNames)):
                data[j].append(self.X_[i][j])
        root = ET.Element('DecisionTree')
        grow_tree(data,categories,root,self.attrNames)
        self.tree_ = ET.tostring(root, encoding="unicode")
        return self

    def predict(self, X):
        check_is_fitted(self, ['tree_', 'resultType', 'attrNames'])
        X = check_array(X)
        dom = minidom.parseString(self.tree_)
        root = dom.childNodes[0]
        prediction = []
        for i in range(len(X)):
            answerlist = decision(root,X[i],self.attrNames,1)
            answerlist = sorted(answerlist.items(), key=lambda x:x[1], reverse = True )
            answer = answerlist[0][0]
            prediction.append((self.resultType)(answer))
        return prediction

    def printTree(self):
        check_is_fitted(self, ['tree_'])
        dom = minidom.parseString(self.tree_)
        print(dom.toprettyxml(newl="\r\n"))

 而后直接按照包的方法进行操作即可得到C4.5算法操作。

# C4.5算法
clf = C45(attrNames=iris.feature_names)
X_train, X_test, y_train, y_test = train_test_split(iris.data, iris.target, test_size=0.5)
clf.fit(X_train, y_train)
print(clf.score(X_test, y_test))