R语言中的机器学习包
来自liuxincumt的博客地址为:
http://blog.csdn.net/liuxincumt/article/details/7527917
Machine Learning & Statistical Learning (机器学习 & 统计学习)
网址:http://cran.r-project.org/web/views/MachineLearning.html维护人员:Torsten Hothorn
版本:2008-02-18 18:19:21
翻译:R-fox, 2008-03-18
机器学习是计算机科学和统计学的边缘交叉领域,R关于机器学习的包主要包括以下几个方面:
1)神经网络(Neural Networks):
nnet包执行单隐层前馈神经网络,nnet是VR包的一部分(http://cran.r-project.org/web/packages/VR/index.html)。
2)递归拆分(Recursive Partitioning):
递归拆分利用树形结构模型,来做回归、分类和生存分析,主要在rpart包(http://cran.r-project.org/web/packages/rpart/index.html)和tree包(http://cran.r-project.org/web/packages/tree/index.html)里执行,尤其推荐rpart包。Weka里也有这样的递归拆分法,如:J4.8, C4.5, M5,包Rweka提供了R与Weka的函数的接口(http://cran.r-project.org/web/packages/RWeka/index.html)。
party包提供两类递归拆分算法,能做到无偏的变量选择和停止标准:函数ctree()用非参条件推断法检测自变量和因变量的关系;而函数mob()能用来建立参数模型(http://cran.r-project.org/web/packages/party/index.html)。另外,party包里也提供二分支树和节点分布的可视化展示。
mvpart包是rpart的改进包,处理多元因变量的问题(http://cran.r-project.org/web/packages/mvpart/index.html)。rpart.permutation包用置换法(permutation)评估树的有效性(http://cran.r-project.org/web/packages/rpart.permutation/index.html)。knnTree包建立一个分类树,每个叶子节点是一个knn分类器(http://cran.r-project.org/web/packages/knnTree/index.html)。LogicReg包做逻辑回归分析,针对大多数自变量是二元变量的情况(http://cran.r-project.org/web/packages/LogicReg/index.html)。maptree包(http://cran.r-project.org/web/packages/maptree/index.html)和pinktoe包(http://cran.r-project.org/web/packages/pinktoe/index.html)提供树结构的可视化函数。
3)随机森林(Random Forests):
randomForest 包提供了用随机森林做回归和分类的函数(http://cran.r-project.org/web/packages/randomForest/index.html)。ipred包用bagging的思想做回归,分类和生存分析,组合多个模型(http://cran.r-project.org/web/packages/ipred/index.html)。party包也提供了基于条件推断树的随机森林法(http://cran.r-project.org/web/packages/party/index.html)。varSelRF包用随机森林法做变量选择(http://cran.r-project.org/web/packages/varSelRF/index.html)。
4)Regularized and Shrinkage Methods:
lasso2包(http://cran.r-project.org/web/packages/lasso2/index.html)和lars包(http://cran.r-project.org/web/packages/lars/index.html)可以执行参数受到某些限制的回归模型。elasticnet包可计算所有的收缩参数(http://cran.r-project.org/web/packages/elasticnet/index.html)。glmpath包可以得到广义线性模型和COX模型的L1 regularization path(http://cran.r-project.org/web/packages/glmpath/index.html)。penalized包执行lasso (L1) 和ridge (L2)惩罚回归模型(penalized regression models)(http://cran.r-project.org/web/packages/penalized/index.html)。pamr包执行缩小重心分类法(shrunken centroids classifier)(http://cran.r-project.org/web/packages/pamr/index.html)。earth包可做多元自适应样条回归(multivariate adaptive regression splines)(http://cran.r-project.org/web/packages/earth/index.html)。
5)Boosting :
gbm包(http://cran.r-project.org/web/packages/gbm/index.html)和boost包(http://cran.r-project.org/web/packages/boost/index.html)执行多种多样的梯度boosting算法,gbm包做基于树的梯度下降boosting,boost包包括LogitBoost和L2Boost。GAMMoost包提供基于boosting的广义相加模型(generalized additive models)的程序(http://cran.r-project.org/web/packages/GAMMoost/index.html)。mboost包做基于模型的boosting(http://cran.r-project.org/web/packages/mboost/index.html)。
6)支持向量机(Support Vector Machines):
e1071包的svm()函数提供R和LIBSVM的接口 (http://cran.r-project.org/web/packages/e1071/index.html)。kernlab包为基于核函数的学习方法提供了一个灵活的框架,包括SVM、RVM……(http://cran.r-project.org/web/packages/kernlab/index.html) 。klaR 包提供了R和SVMlight的接口(http://cran.r-project.org/web/packages/klaR/index.html)。
7)贝叶斯方法(Bayesian Methods):
BayesTree包执行Bayesian Additive Regression Trees (BART)算法(http://cran.r-project.org/web/packages/BayesTree/index.html,http://www-stat.wharton.upenn.edu/~edgeorge/Research_papers/BART%206--06.pdf)。tgp包做Bayesian半参数非线性回归(Bayesian nonstationary, semiparametric nonlinear regression)(http://cran.r-project.org/web/packages/tgp/index.html)。
8)基于遗传算法的最优化(Optimization using Genetic Algorithms):
gafit包(http://cran.r-project.org/web/packages/gafit/index.html)和rgenoud包(http://cran.r-project.org/web/packages/rgenoud/index.html)提供基于遗传算法的最优化程序。
9)关联规则(Association Rules):
arules包提供了有效处理稀疏二元数据的数据结构,而且提供函数执Apriori和Eclat算法挖掘频繁项集、最大频繁项集、闭频繁项集和关联规则(http://cran.r-project.org/web/packages/arules/index.html)。
10)模型选择和确认(Model selection and validation):
e1071包的tune()函数在指定的范围内选取合适的参数(http://cran.r-project.org/web/packages/e1071/index.html)。ipred包的errorest()函数用重抽样的方法(交叉验证,bootstrap)估计分类错误率(http://cran.r-project.org/web/packages/ipred/index.html)。svmpath包里的函数可用来选取支持向量机的cost参数C(http://cran.r-project.org/web/packages/svmpath/index.html)。ROCR包提供了可视化分类器执行效果的函数,如画ROC曲线(http://cran.r-project.org/web/packages/ROCR/index.html)。caret包供了各种建立预测模型的函数,包括参数选择和重要性量度(http://cran.r-project.org/web/packages/caret/index.html)。caretLSF包(http://cran.r-project.org/web/packages/caretLSF/index.html)和caretNWS(http://cran.r-project.org/web/packages/caretNWS/index.html)包提供了与caret包类似的功能。
11)统计学习基础(Elements of Statistical Learning):
书《The Elements of Statistical Learning: Data Mining, Inference, and Prediction 》(http://www-stat.stanford.edu/~tibs/ElemStatLearn/)里的数据集、函数、例子都被打包放在ElemStatLearn包里(http://cran.r-project.org/web/packages/ElemStatLearn/index.html)。