{"id":474,"date":"2023-07-29T19:05:27","date_gmt":"2023-07-29T19:05:27","guid":{"rendered":"https:\/\/statorials.org\/ja\/%e5%a4%9a%e9%a0%85%e5%bc%8f%e5%9b%9e%e5%b8%b0-r\/"},"modified":"2023-07-29T19:05:27","modified_gmt":"2023-07-29T19:05:27","slug":"%e5%a4%9a%e9%a0%85%e5%bc%8f%e5%9b%9e%e5%b8%b0-r","status":"publish","type":"post","link":"https:\/\/statorials.org\/ja\/%e5%a4%9a%e9%a0%85%e5%bc%8f%e5%9b%9e%e5%b8%b0-r\/","title":{"rendered":"R \u306e\u591a\u9805\u5f0f\u56de\u5e30 (\u30b9\u30c6\u30c3\u30d7\u30d0\u30a4\u30b9\u30c6\u30c3\u30d7)"},"content":{"rendered":"

<\/p>\n

\n

\u591a\u9805\u5f0f\u56de\u5e30\u306f\u3001<\/a>\u4e88\u6e2c\u5909\u6570\u3068\u5fdc\u7b54\u5909\u6570<\/a>\u306e\u9593\u306e\u95a2\u4fc2\u304c\u975e\u7dda\u5f62\u3067\u3042\u308b\u5834\u5408\u306b\u4f7f\u7528\u3067\u304d\u308b\u624b\u6cd5\u3067\u3059\u3002<\/span><\/p>\n

\u3053\u306e\u30bf\u30a4\u30d7\u306e\u56de\u5e30\u306f\u6b21\u306e\u5f62\u5f0f\u306b\u306a\u308a\u307e\u3059\u3002<\/span><\/p>\n

Y = \u03b2 0<\/sub> +<\/sup> \u03b2 1<\/sub> X + \u03b2 2<\/sub> X 2<\/sup> + \u2026 + \u03b2 h<\/sub><\/span><\/p>\n

\u3053\u3053\u3067\u3001 h<\/em>\u306f\u591a\u9805\u5f0f\u306e\u300c\u6b21\u6570\u300d\u3067\u3059\u3002<\/span><\/p>\n

\u3053\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u306f\u3001R \u3067\u591a\u9805\u5f0f\u56de\u5e30\u3092\u5b9f\u884c\u3059\u308b\u65b9\u6cd5\u306e\u6bb5\u968e\u7684\u306a\u4f8b\u3092\u793a\u3057\u307e\u3059\u3002<\/span><\/p>\n

\u30b9\u30c6\u30c3\u30d7 1: \u30c7\u30fc\u30bf\u3092\u4f5c\u6210\u3059\u308b<\/strong><\/span><\/h3>\n

\u3053\u306e\u4f8b\u3067\u306f\u300150 \u4eba\u306e\u751f\u5f92\u306e\u30af\u30e9\u30b9\u306e\u5b66\u7fd2\u6642\u9593\u6570\u3068\u6700\u7d42\u8a66\u9a13\u306e\u6210\u7e3e\u3092\u542b\u3080\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f5c\u6210\u3057\u307e\u3059\u3002<\/span><\/p>\n

 #make this example reproducible<\/span>\nset.seed(1)\n\n#create dataset\n<\/span>df <- data.frame(hours = runif<\/span> (50, 5, 15), score=50)\ndf$score = df$score + df$hours^3\/150 + df$hours* runif<\/span> (50, 1, 2)\n\n#view first six rows of data\n<\/span>head(data)\n\n      hours score\n1 7.655087 64.30191\n2 8.721239 70.65430\n3 10.728534 73.66114\n4 14.082078 86.14630\n5 7.016819 59.81595\n6 13.983897 83.60510\n<\/strong><\/pre>\n
\u30b9\u30c6\u30c3\u30d7 2: \u30c7\u30fc\u30bf\u3092\u8996\u899a\u5316\u3059\u308b<\/strong><\/span><\/h3>\n
\u56de\u5e30\u30e2\u30c7\u30eb\u3092\u30c7\u30fc\u30bf\u306b\u5f53\u3066\u306f\u3081\u308b\u524d\u306b\u3001\u307e\u305a\u6563\u5e03\u56f3\u3092\u4f5c\u6210\u3057\u3066\u3001\u5b66\u7fd2\u6642\u9593\u3068\u8a66\u9a13\u30b9\u30b3\u30a2\u306e\u95a2\u4fc2\u3092\u8996\u899a\u5316\u3057\u307e\u3057\u3087\u3046\u3002<\/span> <\/p>\n
 library<\/span> (ggplot2)\n\nggplot(df, aes<\/span> (x=hours, y=score)) +\n  geom_point()<\/strong> <\/pre>\n
\"\"<\/p>\n
\u30c7\u30fc\u30bf\u306b\u306f\u308f\u305a\u304b\u306b 2 \u6b21\u306e\u95a2\u4fc2\u304c\u3042\u308b\u3053\u3068\u304c\u308f\u304b\u308a\u3001\u5358\u7d14\u306a\u7dda\u5f62\u56de\u5e30\u3088\u308a\u3082\u591a\u9805\u5f0f\u56de\u5e30\u306e\u65b9\u304c\u30c7\u30fc\u30bf\u306b\u3088\u304f\u9069\u5408\u3059\u308b\u53ef\u80fd\u6027\u304c\u3042\u308b\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002<\/span><\/p>\n
\u30b9\u30c6\u30c3\u30d7 3: \u591a\u9805\u5f0f\u56de\u5e30\u30e2\u30c7\u30eb\u3092\u5f53\u3066\u306f\u3081\u308b<\/span><\/strong><\/h3>\n
\u6b21\u306b\u3001\u6b21\u6570h<\/em> = 1\u20265 \u3067 5 \u3064\u306e\u7570\u306a\u308b\u591a\u9805\u5f0f\u56de\u5e30\u30e2\u30c7\u30eb\u3092\u8fd1\u4f3c\u3057\u3001k = 10 \u56de\u306e k \u5206\u5272\u76f8\u4e92\u691c\u8a3c\u3092\u4f7f\u7528\u3057\u3066\u5404\u30e2\u30c7\u30eb\u306e MSE \u691c\u5b9a\u3092\u8a08\u7b97\u3057\u307e\u3059\u3002<\/span><\/p>\n
 #randomly shuffle data\n<\/span>df.shuffled <- df[ sample<\/span> ( nrow<\/span> (df)),]\n\n#define number of folds to use for k-fold cross-validation\n<\/span>K <- 10 \n\n#define degree of polynomials to fit\n<\/span>degree <- 5\n\n#create k equal-sized folds\n<\/span>folds <- cut( seq<\/span> (1, nrow<\/span> (df.shuffled)), breaks=K, labels= FALSE<\/span> )\n\n#create object to hold MSE's of models\n<\/span>mse = matrix(data=NA,nrow=K,ncol=degree)\n\n#Perform K-fold cross validation\n<\/span>for<\/span> (i in<\/span> 1:K){\n    \n#define training and testing data\n<\/span>testIndexes <- which<\/span> (folds==i,arr.ind= TRUE<\/span> )\n    testData <- df.shuffled[testIndexes, ]\n    trainData <- df.shuffled[-testIndexes, ]\n    \n#use k-fold cv to evaluate models\n<\/span>for (j in 1:degree){\n        fit.train = lm<\/span> (score ~ poly<\/span> (hours,d), data=trainData)\n        fit.test = predict<\/span> (fit.train, newdata=testData)\n        mse[i,j] = mean<\/span> ((fit.test-testData$score)^2) \n    }\n}\n\n#find MSE for each degree \n<\/span>colMeans(mse)\n\n[1] 9.802397 8.748666 9.601865 10.592569 13.545547\n<\/strong><\/span><\/pre>\n
\u7d50\u679c\u304b\u3089\u3001\u5404\u30e2\u30c7\u30eb\u306e MSE \u30c6\u30b9\u30c8\u304c\u308f\u304b\u308a\u307e\u3059\u3002<\/span><\/p>\n