The Caret Package

来源：互联网发布：淘宝当季新品怎么设置编辑：程序博客网时间：2024/05/16 04:07

今天再用逻辑斯特回归做模式识别作业,忽然发现了神经网络包 ,遂看了一下持续更新

#载入DMwR包library(DMwR) #载入nnet(神经网络)包library(nnet) #利用read.table读取数据original.data = read.table("E:/MyDownloads/dataset/dataset/test.data",header = FALSE,sep = " ")#给第11列命名为“label"colnames(original.data)[11] ="label"#for(i in 1:nrow(original.data)){#  if(original.data[i,"label"] == 1)#    original.data[i,"label"] = "a"#  if(original.data[i,"label"] == -1)#    original.data[i,"label"] = "b"#}#调用神经网络函数 nn <- nnet(label~.,original.data, size = 10, decay = 0.01,            maxit = 1000, linout = T, trace = F) norm.preds <- predict(nn, original.data) norm.preds[which(norm.preds>0)] = 1norm.preds[which(norm.preds<0)] = -1a =table(norm.preds,original.data$label) head(a)#linout 默认是逻辑输出, T为线性输出#maixt 最大迭代次数#size 隐层数量

caret包是一个功能很强大的包，有通用接口函数train()，用来调用各种训练方法我们以’nnet ‘网络为例

library(caret)#载入DMwR包library(DMwR) #载入nnet(神经网络)包library(nnet) traindata 可以自己使用iris数据#get index for seperate data.#split dataset automatically.trainIndex <- createDataPartition(traindata$a, p = .8,                                   list = FALSE,                                   times = 1)head(trainIndex)#as far as I thought data should not have factor variables#tune parameters my.grid <- expand.grid(.decay = c(0,0.5, 0.1), .size = c(4, 5, 6, 7))#nnetfit <- train(y~x, train.dat,              method = 'nnet', linout = TRUE,             trace  = FALSE,             tuneGrid = my.grid)#when predict, newdata feature label should be the same as data, except for label.(重要，不然会出错）pred_y <- predict(fit$finalModel, newdata = data.frame(x = test.dat$x))

caret 包功能强大各种数据处理， cross validation, sampling, data preprocess.
机器学习比赛用R语言的话最好把 caret包学习一下
用的到的话推荐网址caret learning tutor

0 0