关于回归方法,请参考我的另一篇博客30分钟学会用scikit-learn的基本回归方法(线性、决策树、SVM、KNN)和集成方法(随机森林,Adaboost和GBRT)
本文主要参考了scikit-learn的官方网站
1. 数据准备
关于分类,我们使用了Iris数据集,这个scikit-learn自带了.
Iris数据集是常用的分类实验数据集,由Fisher, 1936收集整理。Iris也称鸢尾花卉数据集,是一类多重变量分析的数据集。数据集包含150个数据集,分为3类,每类50个数据,每个数据包含4个属性。可通过花萼长度,花萼宽度,花瓣长度,花瓣宽度4个属性预测鸢尾花卉属于(Setosa,Versicolour,Virginica)三个种类中的哪一类。
注意,Iris数据集给出的三种花是按照顺序来的,前50个是第0类,51-100是第1类,101~150是第二类,如果我们分训练集和测试集的时候要把顺序打乱
这里我们引入一个两类shuffle的函数,它接收两个参数,分别是x和y,然后把x,y绑在一起shuffle.
def shuffle_in_unison(a, b):
assert len(a) == len(b)
import numpy
shuffled_a = numpy.empty(a.shape, dtype=a.dtype)
shuffled_b = numpy.empty(b.shape, dtype=b.dtype)
permutation = numpy.random.permutation(len(a))
for old_index, new_index in enumerate(permutation):
shuffled_a[new_index] = a[old_index]
shuffled_b[new_index] = b[old_index]
return shuffled_a, shuffled_b
下面我们导入Iris数据并打乱它,然后分为100个训练集和50个测试集
from sklearn.datasets import load_iris
iris = load_iris()
def load_data():
iris.data, iris.target = shuffle_in_unison(iris.data, iris.target)
x_train ,x_test = iris.data[:100],iris.data[100:]
y_train, y_test = iris.target[:100].reshape(-1,1),iris.target[100:].reshape(-1,1)
return x_train, y_train, x_test, y_test
2. 试验各种不同的方法
常用的分类方法一般有决策树, SVM, kNN, 朴素贝叶斯, 集成方法有随机森林,Adaboost和GBDT
完整代码如下:
from sklearn.datasets import load_iris
iris = load_iris()
def shuffle_in_unison(a, b):
assert len(a) == len(b)
import numpy
shuffled_a = numpy.empty(a.shape, dtype=a.dtype)
shuffled_b = numpy.empty(b.shape, dtype=b.dtype)
permutation = numpy.random.permutation(len(a))
for old_index, new_index in enumerate(permutation):
shuffled_a[new_index] = a[old_index]
shuffled_b[new_index] = b[old_index]
return shuffled_a, shuffled_b
def load_data():
iris.data, iris.target = shuffle_in_unison(iris.data, iris.target)
x_train ,x_test = iris.data[:100],iris.data[100:]
y_train, y_test = iris.target[:100].reshape(-1,1),iris.target[100:].reshape(-1,1)
return x_train, y_train, x_test, y_test
from sklearn import tree, svm, naive_bayes,neighbors
from sklearn.ensemble import BaggingClassifier, AdaBoostClassifier, RandomForestClassifier, GradientBoostingClassifier
x_train, y_train, x_test, y_test = load_data()
clfs = {'svm': svm.SVC(),\
'decision_tree':tree.DecisionTreeClassifier(),
'naive_gaussian': naive_bayes.GaussianNB(), \
'naive_mul':naive_bayes.MultinomialNB(),\
'K_neighbor' : neighbors.KNeighborsClassifier(),\
'bagging_knn' : BaggingClassifier(neighbors.KNeighborsClassifier(), max_samples=0.5,max_features=0.5), \
'bagging_tree': BaggingClassifier(tree.DecisionTreeClassifier(), max_samples=0.5,max_features=0.5),
'random_forest' : RandomForestClassifier(n_estimators=50),\
'adaboost':AdaBoostClassifier(n_estimators=50),\
'gradient_boost' : GradientBoostingClassifier(n_estimators=50, learning_rate=1.0,max_depth=1, random_state=0)
}
def try_different_method(clf):
clf.fit(x_train,y_train.ravel())
score = clf.score(x_test,y_test.ravel())
print('the score is :', score)
for clf_key in clfs.keys():
print('the classifier is :',clf_key)
clf = clfs[clf_key]
try_different_method(clf)
给出的结果如下:
the classifier is : svm
the score is : 0.94
the classifier is : decision_tree
the score is : 0.88
the classifier is : naive_gaussian
the score is : 0.96
the classifier is : naive_mul
the score is : 0.8
the classifier is : K_neighbor
the score is : 0.94
the classifier is : gradient_boost
the score is : 0.88
the classifier is : adaboost
the score is : 0.62
the classifier is : bagging_tree
the score is : 0.94
the classifier is : bagging_knn
the score is : 0.94
the classifier is : random_forest
the score is : 0.92