{"id":1543,"date":"2018-08-26T12:05:37","date_gmt":"2018-08-26T03:05:37","guid":{"rendered":"https:\/\/plaza.umin.ac.jp\/~OIO\/?p=1543"},"modified":"2018-08-26T14:20:48","modified_gmt":"2018-08-26T05:20:48","slug":"r%e3%81%a7pubmed%e3%81%ae%e6%a4%9c%e7%b4%a2%e7%b5%90%e6%9e%9c%e3%82%92%e6%a9%9f%e6%a2%b0%e5%ad%a6%e7%bf%92","status":"publish","type":"post","link":"https:\/\/plaza.umin.ac.jp\/~OIO\/?p=1543","title":{"rendered":"R\u3067PubMed\u306e\u691c\u7d22\u7d50\u679c\u3092\u6a5f\u68b0\u5b66\u7fd2"},"content":{"rendered":"

\u306f\u3058\u3081\u306b<\/h2>\n

\u672c\u7a3f\u306fR\u306e\u30d1\u30c3\u30b1\u30fc\u30b8\u3067deep learning\u304c\u3067\u304d\u308b\u3068\u805e\u3044\u3066\u3001\u30cd\u30c3\u30c8\u3067\u8abf\u3079\u306a\u304c\u3089\u3001\u30d1\u30c3\u30b1\u30fc\u30b8\u3092\u4f7f\u3046\u9053\u7b4b\u3092\u3064\u3051\u308b\u307e\u3067\u306e\u3001\u5fd8\u5099\u9332\u3067\u3059\u3002\u30b5\u30f3\u30d7\u30eb\u306b\u3057\u305f\u306e\u306f\u3001\u300cA\u85ac\u300d\u3067\u691c\u7d22\u3057\u305f\u7d50\u679c\u3068\u300cB\u85ac\u300d\u3067\u691c\u7d22\u3057\u305f\u7d50\u679c\u3092\u3001\u30c6\u30ad\u30b9\u30c8\u5f62\u5f0f\u3067A.txt, B.txt\u3068\u3057\u3066PC\u306b\u4fdd\u5b58\u3057\u305f\u3046\u3048\u3067\u3001\u4e00\u90e8\u3092\u5b66\u7fd2\u7528\u30b5\u30f3\u30d7\u30eb\u3001\u6b8b\u308a\u3092\u30c6\u30b9\u30c8\u7528\u30b5\u30f3\u30d7\u30eb\u306b\u3057\u3066\u3001\u5b66\u7fd2\u306e\u52b9\u679c\uff08\uff1f\uff09\u3092\u6e2c\u5b9a\u3057\u307e\u3057\u305f\u3002\u96c6\u8a08\u306b\u5165\u308b\u524d\u306b\u3001\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u305f\u30c7\u30fc\u30bf\u306e\u4e2d\u306e\u3001\u691c\u7d22\u30ad\u30fc\u30ef\u30fc\u30c9\u3092\u524a\u9664\u3057\u307e\u3057\u305f\u3002\uff08\u2190\u3053\u308c\u3092\u3057\u306a\u3044\u3068\u3055\u3059\u304c\u306b\u691c\u7d22\u30ad\u30fc\u30ef\u30fc\u30c9\u305d\u306e\u3082\u306e\u304c\u305d\u308c\u305e\u308c\u306e\u96c6\u56e3\u306b\u3070\u3063\u3061\u308a\u5165\u3063\u3066\u3057\u307e\u3046\u306e\u3067\u5206\u985e\u306e\u6027\u80fd\u304c\u8a55\u4fa1\u3067\u304d\u306a\u3044\u306e\u3067\u306f\u306a\u3044\u304b\u3068\uff09<\/p>\n

\"\"<\/p>\n

\u306f\u3063\u304d\u308a\u5206\u985e\u3057\u3084\u3059\u3044\u3088\u3046\u306b\u3001\u4eca\u56de\u306fA\u306f\u6297\u764c\u85ac\u3001B\u306f\u514d\u75ab\u795e\u7d4c\u7cfb\u75be\u60a3\u6cbb\u7642\u85ac\u3068\u6cbb\u7642\u5bfe\u8c61\u306e\u75be\u60a3\u306e\u6027\u8cea\u304c\u9055\u3046\u3082\u306e\u306b\u3057\u3066\u3044\u307e\u3059\u3002\u691c\u7d22\u7d50\u679c\u306f\u3001Medline\u5f62\u5f0f\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u307e\u3059\u3002\u3053\u306e\u3042\u305f\u308a\u306e\u624b\u9806\u306f\u3001\u3010R\u3067\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u3011<\/a>\u3092\u3001\u307b\u307c\u305d\u306e\u307e\u307e\u5229\u7528\u3057\u3066\u3044\u307e\u3059\u3002\uff08\u300c\u30c7\u30fc\u30bf\u5206\u6790\u7cfb\u7537\u5b50\u300d\u3055\u3093\u3042\u308a\u304c\u3068\u3046\u3054\u3056\u3044\u307e\u3059\u3002\uff09\u5206\u985e\u306b\u4f7f\u7528\u3057\u305fMXNet\u306e\u5468\u308a\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u306f\u3010R\u306eMXNet\u3067iris\u3092\u5206\u985e\u3011<\/a>\u3092\u3001\u307b\u307c\u305d\u306e\u307e\u307e\u5229\u7528\u3057\u3066\u3044\u307e\u3059\uff08\u300c\u306a\u3093\u3068\u306a\u304f\u306aDeveloper\u300d\u3055\u3093\u3042\u308a\u304c\u3068\u3046\u3054\u3056\u3044\u307e\u3059\u3002\uff09<\/p>\n

\n

{mxnet}\u30d1\u30c3\u30b1\u30fc\u30b8\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb<\/h1>\n

\u3068\u308a\u3042\u3048\u305a\u3053\u306e\u30d1\u30c3\u30b1\u30fc\u30b8\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3067\u3059\u3002\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u306f\u3001\u4ed6\u306e\u65e5\u672c\u8a9e\u306e\u30b5\u30a4\u30c8\u306e\u8aac\u660e\u901a\u308a\u3067\u306f\u3001\u306a\u305c\u304b\u3046\u307e\u304f\u3086\u304d\u307e\u305b\u3093\u3002\u4e00\u5fdc\u4ee5\u4e0b\u306e\u6d41\u308c\u3067\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3067\u304d\u307e\u3057\u305f\u3002\u4ed6\u306e\u30d1\u30c3\u30b1\u30fc\u30b8\u306f\u7279\u306b\u82e6\u52b4\u3059\u308b\u3053\u3068\u306a\u304finstall.packages()\u3067\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n

# first add the repo
\ndrat::addRepo(“dmlc”)
\n# either install just one or more given packages
\ninstall.packages(“xgboost”)<\/p>\n

cran <- getOption(“repos”)
\ncran[“dmlc”] <- “https:\/\/s3-us-west-2.amazonaws.com\/apache-mxnet\/R\/CRAN\/”
\noptions(repos = cran)
\ninstall.packages(“mxnet”)<\/p>\n

\n<\/blockquote>\n

MedLine\u691c\u7d22\u3068\u4fdd\u7ba1<\/h2>\n

\u691c\u7d22\u30ad\u30fc\u30ef\u30fc\u30c9\u306f\u666e\u901a\u306b\u5165\u529b\u3057\u3066\u691c\u7d22\u3059\u308c\u3070\u3088\u3044\u306e\u3067\u3059\u304c\u3001\u3042\u307e\u308a\u305f\u304f\u3055\u3093\u306e\u6587\u732e\u304c\u3042\u3063\u3066\u3082\u53d6\u308a\u56de\u3057\u304c\u60aa\u3044\u306e\u3067\u4eca\u56de\u306f\u691c\u7d22\u5bfe\u8c61\u3092\u4eca\u5e74\u51fa\u7248\u306e\u6587\u732e\u3068\u3057\u307e\u3057\u305f\u3002\u307e\u305f\u30a2\u30d6\u30b9\u30c8\u30e9\u30af\u30c8\u3067\u5206\u985e\u3057\u307e\u3059\u306e\u3067\u3001\u30a2\u30d6\u30b9\u30c8\u30e9\u30af\u30c8\u304c\u5165\u529b\u3055\u308c\u3066\u3044\u308b\u82f1\u8a9e\u306e\u6587\u732e\u306b\u7d5e\u3063\u3066\u51fa\u529b\u3059\u308b\u3088\u3046\u306b\u3057\u307e\u3057\u305f\u3002\"\"<\/p>\n

\u4fdd\u5b58\u306b\u5f53\u305f\u3063\u3066\u306f\u3001send to\u304b\u3089File-Medline\u5f62\u5f0f\u3092\u9078\u629e\u3057\u307e\u3057\u305f\u3002<\/p>\n

\"\"<\/p>\n

Medline\u8aad\u307f\u8fbc\u307f<\/h2>\n

setwd(“C:\/Users\/Oshima\/Documents\/2018\/R deep learning\/TXT\/”)<\/p>\n

## define function
\nmedline <- function(file_name){
\nlines <- readLines(file_name)
\nmedline_records <- list()
\nkey <- 0
\nrecord <- 0
\nfor(line in lines){
\nheader <- sub(” {1,20}”, “”, substring(line, 1, 4))
\nvalue <- sub(“^.{6}”, “”, line)
\nif(header == “” & value == “”){
\nnext
\n}
\nelse if(header == “PMID”){
\nrecord = record + 1
\nmedline_records[[record]] <- list()
\nmedline_records[[record]][header] <- value
\n}
\nelse if(header == “” & value != “”){
\nmedline_records[[record]][key] <- paste(medline_records[[record]][key], value)
\n}
\nelse{
\nkey <- header
\nif(is.null(medline_records[[record]][key][[1]])){
\nmedline_records[[record]][key] <- value
\n}
\nelse {
\nmedline_records[[record]][key] <- paste(medline_records[[record]][key], value, sep=”;”)
\n}
\n}
\n}
\nreturn(medline_records)
\n}<\/p>\n

## read Medline
\nfileVec <- list.files(file.path(getwd()),
\npattern=”.txt”,
\nfull.names = T)
\ncategoryVec <- list.files(file.path(getwd()),
\npattern=”.txt”)
\ndataMed <- lapply(fileVec,medline)<\/p>\n

## exstract Title, Abstract and CategoryVec
\ndataMedList <- lapply(seq(1,length(dataMed)),function(x){
\nres <- lapply(seq(1,length(dataMed[[x]])),function(y){
\nout <- data.frame(title=dataMed[[x]][[y]]$TI,
\nabst=dataMed[[x]][[y]]$AB,
\ncategory=categoryVec[x])
\nreturn(out)
\n# return(x)
\n})
\nreturn(do.call(rbind,res))
\n# return(y)
\n})<\/p>\n

dataMedDF <- do.call(rbind,dataMedList)<\/p><\/blockquote>\n

\u30c7\u30fc\u30bf\u306e\u524d\u51e6\u7406<\/h2>\n

dataMH <- sapply(dataMed,function(i){
\nsapply(i,function(x){
\nif(all(!(names(x) %in% “MH”))){
\nstrMH <- c()
\n}else{
\nstrMH <- strsplit(x$MH,”;”)
\n}
\nstrMHgsub <- gsub(” “,”_”,strMH[[1]])
\nstrMHpaste <- paste(strMHgsub,collapse = ” “)
\nreturn(strMHpaste)
\n})
\n})<\/p><\/blockquote>\n

\u30c8\u30d4\u30c3\u30af\u983b\u5ea6\u306e\u30c7\u30fc\u30bf\u30d5\u30ec\u30fc\u30e0\u3092\u4f5c\u6210\u3057\u307e\u3057\u305f\u3002\u3053\u3053\u3067\u306f\u3001\u4e0a\u8ff0\u306e\u30b5\u30a4\u30c8\u306b\u5023\u3063\u3066k=20\u3067\u3084\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n

library(textmineR)
\nlibrary(text2vec)
\nlibrary(tm)
\nlibrary(topicmodels)<\/p>\n

# create vector of abstracts
\nallTexts <- sapply(dataMed,function(i){
\nsapply(i,function(x){
\nti <- gsub(“\\\\[|\\\\]”,””,x$TI)
\nif(all(!(names(x) %in% “MH”))){
\nstrMH <- c()
\n}else{
\nstrMH <- strsplit(x$MH,”;”)
\n}
\nstrMHgsub <- gsub(” “,”_”,strMH[[1]])
\nstrMHpaste <- paste(strMHgsub,collapse = ” “)
\npaste(ti,x$AB,strMHpaste,collapse=” “)
\n})
\n})<\/p>\n

allTexts <- unlist(allTexts)<\/p>\n

# preprocess
\nsw <- c(“i”, “me”, “my”, “myself”, “we”, “our”, “ours”,
\n“ourselves”, “you”, “your”, “yours”, tm:::stopwords(“English”))
\npreText <- tolower(allTexts)
\npreText <- tm::removePunctuation(preText)
\npreText <- tm::removeWords(preText,sw)
\npreText <- tm::removeNumbers(preText)
\npreText <- tm::stemDocument(preText, language = “english”)<\/p>\n

# tokenize(split into single words)
\nit <- itoken(preText,
\npreprocess_function = tolower,
\ntokenizer = word_tokenizer)
\n#, ids = abstID)<\/p>\n

# delete stopwords and build vocablary
\nvocab <- create_vocabulary(it,
\nstopwords = sw)<\/p>\n

# word vectorize
\nvectorizer <- vocab_vectorizer(vocab)<\/p>\n

# create DTM
\ndtm <- create_dtm(it, vectorizer)
\n# Term frecency
\nTDF <- TermDocFreq(dtm)<\/p>\n

# modeling by package “topicmodels”
\nmodel <- LDA(dtm, control=list(seed=37464847),k = 20, method = “Gibbs”)<\/p>\n

termsDF <- get_terms(model,100)
\ntopicProbability <- data.frame(model@gamma,dataMedDF$category)<\/p><\/blockquote>\n

\u4eca\u56de\u306f\u5b66\u7fd2\u7528\u30b5\u30f3\u30d7\u30eb\u309280%, \u8a55\u4fa1\u7528\u30b5\u30f3\u30d7\u30eb\u309220%\u306b\u3057\u3066\u3001\u5206\u3051\u307e\u3057\u305f\u3002<\/p>\n

train_size <- 0.8
\nn <- nrow(topicProbability)
\nperm <- sample(n, size=round(n * train_size))<\/p>\n

# data for training
\ntrain <- topicProbability[perm, ]<\/p>\n

# data for test
\ntest <- topicProbability[-perm, ]<\/p>\n

# imput data for training
\ntrain.x <- data.matrix(train[1:20])<\/p>\n

# label for training
\ntrain.y <- as.numeric(train$dataMedDF.category) -1<\/p>\n

# imput data for test
\ntest.x <- data.matrix(test[1:20])<\/p>\n

# label for test
\ntest.y <- as.numeric(test$dataMedDF.category) -1<\/p><\/blockquote>\n

\n

\u4ee5\u4e0a\u30c7\u30fc\u30bf\u306e\u524d\u51e6\u7406\u3067\u3057\u305f\u3002<\/p>\n

\u968e\u5c64\u578b\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af<\/h2>\n

\u30c7\u30fc\u30bf\u306e\u524d\u51e6\u7406\u304c\u7d42\u308f\u308a\u307e\u3057\u305f\u306e\u3067\u3001\u3053\u3053\u304b\u3089\u304c\u3044\u308f\u3086\u308b\u6a5f\u68b0\u5b66\u7fd2\u306e\u51e6\u7406\u306b\u306a\u308a\u307e\u3059\u3002iris\u8a18\u4e8b\u306b\u5023\u3063\u3066\u3001\u672c\u7a3f\u3067\u3082\u6b21\u306e\u3088\u3046\u306b\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n\n\t\n\n\t\n\t\n\t\n\t\n\t\n\t
\u5f15\u6570<\/th>\u5099\u8003<\/th>\u4eca\u56de\u306e\u30d1\u30e9\u30e1\u30fc\u30bf<\/th>\n<\/tr>\n<\/thead>\n
hidden_node<\/td>\u96a0\u308c\u5c64\u306e\u30ce\u30fc\u30c9\uff08\u30cb\u30e5\u30fc\u30ed\u30f3\uff09\u6570\u3000\u30c7\u30d5\u30a9\u30eb\u30c8\u306f\uff11<\/td>5<\/td>\n<\/tr>\n
out_node<\/td>\u51fa\u529b\u30ce\u30fc\u30c9\u6570\uff08\u5206\u985e\u6570\u3001\u4eca\u56de\u306fA\u85ac\u3068B\u85ac\u306e\uff12\uff09<\/td>2<\/td>\n<\/tr>\n
num.round<\/td>\u7e70\u308a\u8fd4\u3057\u56de\u6570\uff08\u30c7\u30d5\u30a9\u30eb\u30c8\u306f\uff11\uff10\uff09<\/td>100<\/td>\n<\/tr>\n
array.batch.size<\/td>\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba\uff08\u30c7\u30d5\u30a9\u30eb\u30c8\u306f128\uff09<\/td>10<\/td>\n<\/tr>\n
learning.rate<\/td>\u5b66\u7fd2\u4fc2\u6570<\/td>0.1<\/td>\n<\/tr>\n
activation<\/td>\u6d3b\u6027\u5316\u4fc2\u6570\uff08\u30c7\u30d5\u30a9\u30eb\u30c8\u306f'tanh'\uff09<\/td>'relu'<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n

‘tanh’, ‘relu’\u3068\u306f\u306a\u3093\u305e\u3084\u3001\u3068\u3044\u3046\u306e\u306f\u89e3\u3089\u306a\u3044\u3002\u300c\u30d5\u30ea\u30fc\u30e9\u30f3\u30b9\u306e\u30d7\u30ed\u30b0\u30e9\u30de\u300d\u3055\u3093\u306b\u3088\u308b\u3068<\/a>\u3001\u300c\u7d50\u8ad6\u304b\u3089\u8a00\u3046\u3068Relu\u3092\u4f7f\u304a\u3046\u300d\u306a\u306e\u3060\u305d\u3046\u3067\u3059\u3002<\/p>\n

# deep learning<\/p>\n

library(mxnet)
\nmx.set.seed(0)<\/p>\n

# \u5b66\u7fd2
\nmodel <- mx.mlp(train.x, train.y,
\nhidden_node = 5,
\nout_node = 2,
\nnum.round = 100,
\nlearning.rate = 0.1,
\narray.batch.size = 10,
\nactivation = ‘relu’,
\narray.layout = ‘rowmajor’,
\neval.metric = mx.metric.accuracy)<\/p>\n

# \u8a55\u4fa1
\npred <- predict(model, test.x, array.layout = ‘rowmajor’)<\/p>\n

# \u8a55\u4fa1\u7528\u30c7\u30fc\u30bf\u306e\u5206\u985e\u7d50\u679c\uff080, 1\uff09
\npred.y <- max.col(t(pred)) -1<\/p>\n

# \u8a55\u4fa1\u30c7\u30fc\u30bf\u306e\u6b63\u89e3\u306e\u5272\u5408\u3092\u7b97\u51fa
\nacc <- sum(pred.y == test.y) \/ length(pred.y)<\/p>\n

print(acc)<\/p><\/blockquote>\n

\u7d50\u679c<\/h2>\n

\u6b21\u306e\u901a\u308a97.1% (95%CI; 92.6 – 99.2) \u306e\u5272\u5408\u3067\u6b63\u89e3\u306b\u5206\u985e\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n

\"\"<\/p>\n","protected":false},"excerpt":{"rendered":"

\u306f\u3058\u3081\u306b \u672c\u7a3f\u306fR\u306e\u30d1\u30c3\u30b1\u30fc\u30b8\u3067deep learning\u304c\u3067\u304d\u308b\u3068\u805e\u3044\u3066\u3001\u30cd\u30c3\u30c8\u3067\u8abf\u3079\u306a\u304c\u3089\u3001\u30d1\u30c3\u30b1\u30fc\u30b8\u3092\u4f7f\u3046\u9053\u7b4b\u3092\u3064\u3051\u308b\u307e\u3067\u306e\u3001\u5fd8\u5099\u9332\u3067\u3059\u3002\u30b5\u30f3\u30d7\u30eb\u306b\u3057\u305f\u306e\u306f\u3001\u300cA\u85ac\u300d\u3067\u691c\u7d22\u3057\u305f\u7d50\u679c\u3068\u300cB\u85ac\u300d\u3067\u691c\u7d22\u3057\u305f\u7d50\u679c\u3092\u3001\u30c6\u30ad\u30b9…<\/p>\n","protected":false},"author":1,"featured_media":1565,"comment_status":"open","ping_status":"open","sticky":true,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[4],"tags":[],"class_list":["post-1543","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-science"],"jetpack_featured_media_url":"https:\/\/plaza.umin.ac.jp\/~OIO\/wp-content\/uploads\/2018\/08\/deepLearning.png","jetpack_shortlink":"https:\/\/wp.me\/p9b6zl-oT","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/posts\/1543","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1543"}],"version-history":[{"count":0,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/posts\/1543\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=\/wp\/v2\/media\/1565"}],"wp:attachment":[{"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1543"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1543"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/plaza.umin.ac.jp\/~OIO\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1543"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}