ããã¯ãã£ã¹ãã¬ã¤åºåã«éã£ã話ã§ã¯ãªãã¨æããã§ãããããPC / ã¹ããä¸ã®ä½ãããã®ã¯ãªã¨ã¤ãã£ãã«å¯¾ããã¯ãªãã¯æ°ããã®ãã¶ã¤ã³ã®è¯ãæªãã«ãã£ã¦å·¦å³ãããã¨ãããã¨ã¯ãwebæ¥çã§ã¯ããç¥ããã¦ãããã¨æãã¾ãã
ããããå ´åãã©ããªãã¶ã¤ã³ãã¯ãªãã¯æ°ãå¢ããã®ã«æå¹ãï¼ãã¨ããã®ã¯ãå³å¯ã«ã¯ãã¡ãã¨æ¡ä»¶çµ±å¶ãããã¦å®é¨è¨ç»æ³ã«åºã¥ãã¦ãã¶ã¤ã³ããA/Bãã¹ããªã©ã§èª¿ã¹ãã¹ããªãã§ãããããããã¾ã§ç¶¿å¯ã«ãã£ã¦ããä½è£ã®ãªãç¾å ´ãçµæ§å¤ãã¯ãã
ããã§ãä»åã¯æ¢ã«è¨æ¸¬æ¸ã¿ã®ååºåã®ã¯ãªãã¯æ°ï¼CTRã§ããå¾ããã¦ããªãããã§ããã°å®ã¯ãªãã¯æ°ã«ç´ããã®ã¨æ³å®ããï¼ãã¼ã¿ãå¾ããã¦ãããã®ã¨ä»®å®ãã¦ããããååºåã®ãã¶ã¤ã³è¦ç´ ã表ãã¤ã³ããã¯ã¹ï¼äºå¤orã«ãã´ãªã«ã«ãã¼ã¿ï¼ã®ãã¼ã¿ã¨çµã¿åããã¦ããã©ããªãã¶ã¤ã³ãããã°ã¯ãªãã¯æ°ãå¢ãããï¼ããæ¨å®ããã¨ããã±ã¼ã¹ãæ³å®ãã¦Rã§ã©ãããã°è¯ãããèãã¦ã¿ããã¨æãã¾ãã
ã¡ãªã¿ã«ãä»åç¨ãããã¢ã½ã³å帰ã¢ãã«ãåãã¨ããGLMå¨ãã®åºç¤ç¥èã«ã¤ãã¦ã¯ãã¡ãã®æ¬ãæãåãããããã¨æãã¾ããæ¸è©ãéä¸ã¾ã§æ¸ãã¦ãæãå¤ãã¦çµå±æ¸ãã®ãããã¦ãã¾ã£ããã§ãããåãããããã§ã¯ä»ã®è¿½éã許ããªãåèã§ãã
- ä½è : ä¹ ä¿æå¼¥
- åºç社/ã¡ã¼ã«ã¼: 岩波æ¸åº
- çºå£²æ¥: 2012/05/19
- ã¡ãã£ã¢: åè¡æ¬
- è³¼å ¥: 16人 ã¯ãªãã¯: 163å
- ãã®ååãå«ãããã° (17件) ãè¦ã
ã¾ããåãä¹ ä¿å çã®æã«ãã解説ãã¼ã¸ï¼生態学データ解析 - FAQ 一般化線形モデルï¼ã大ãã«åèã«ãªãã®ã§ã¯ãªããã¨ã
å¿ è¦ãªRããã±ã¼ã¸
{MASS}ã¨{randomForest}ãã¤ã³ã¹ãã¼ã«ãã¦å±éãã¦ä¸ãããããã¿ã¤ãã«ã«ããã¨æ¸ããªãã£ãã®ãããã§ãã¬ã¡ããã¾ãããï¼ç¬ï¼ã
ãã¢ã½ã³åå¸ã«ããGLMï¼ãã¢ã½ã³å帰ã¢ãã«ï¼ã§ãã£ã¦ã¿ã
ãã¤ãéãサンプルをGitHubに上げておきましたã®ã§è½ã¨ãã¦ãã¦ãã ãã*1ãååã¯ä½ã§ããããã§ãããä»åã¯é©å½ã«dat1ã¨ããã¦ããã¾ãã
ãã¼ã¿ã®ã¤ã¡ã¼ã¸ã¨ãã¦ã¯ããã£ã¹ãã¬ã¤åºåã®ãã¶ã¤ã³è¦ç´ ã12種é¡ã«åãã¦ãããããã®ãããorãªããã®çµã¿åããã§åã ã®åºåã表ç¾ãã¦ãããã®ã¨ãã¾ãï¼d1-d12ã®åã«ã©ã ã§ãããããªã1ï¼ããªãããªã0ï¼ãããã¦ãããããã®åºåã®ã¯ãªãã¯æ°ã®å®æ°ãæå¾ã«è¨é²ããã¦ãããã®ã¨ãã¾ãï¼clickã«ã©ã ï¼ãããã«ããããã¡ãªãã¼ã¿ã§ããã
d1 | d2 | d3 | d4 | d5 | d6 | d7 | d8 | d9 | d10 | d11 | d12 | click |
---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 1 | 1 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 175 |
0 | 1 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 350 |
1 | 1 | 1 | 0 | 1 | 0 | 1 | 1 | 0 | 1 | 1 | 0 | 675 |
0 | 1 | 0 | 1 | 1 | 0 | 1 | 1 | 0 | 0 | 1 | 0 | 250 |
0 | 1 | 0 | 1 | 1 | 1 | 1 | 1 | 0 | 1 | 1 | 1 | 275 |
... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
ã¾ãã¯ããç®çå¤æ°ï¼ã«ã¦ã³ããã¼ã¿ / 説æå¤æ°ï¼äºå¤ãã¼ã¿ãã¨ãããã¨ã§ã»ãªãªã¼éããã¢ã½ã³å帰ã¢ãã«ã§ãã£ã¦ã¿ããã¨æãã¾ãããã®å帰å¼ã¯ã説æå¤æ°ããç®çå¤æ°ãã¨ããã¨ä»¥ä¸ã®éãã
è¦ã¯ããã¢ã½ã³åå¸ã®ãã©ã¡ã¼ã¿Î»ãç·å½¢ã¢ãã«ã§èª¬æå¯è½ãªãã®ã¨ä»®å®ãã¦ããã®ç·å½¢ã¢ãã«ãæ¨å®ããç®çå¤æ°ã対æ°å¤æãããã®ã«å¯¾ãã¦ãé常ã®ç·å½¢ï¼éï¼å帰ã¢ãã«ãæ¨å®ãã*2ãã®ã§ãããã®è¾ºã®çå±ã¯ä¾ãã°こちらのページã¨ãããããã¯id:uncorrelatedããã®ブログ記事ãåç
§ã®ãã¨ï¼ãæææé£ããããã¾ããï¼ãã¡ãªã¿ã«ç·å½¢å帰ã¢ãã«ã®åºç¤ã«ã¤ãã¦ã¯以前の記事ã§ã¡ãã£ã¨è§¦ãã¦ããã®ã§ããã¡ãããèªã¿ããã
Rã§ã®è¨ç®èªä½ã¯ãglm(){stats}ã§family=poissonã¨ããã°ç°¡åã«å®è¡ã§ãã¾ãã
> dat1.glm<-glm(click~.,dat1,family=poisson) > summary(dat1.glm) Call: glm(formula = click ~ ., family = poisson, data = dat1) Deviance Residuals: Min 1Q Median 3Q Max -23.149 -9.143 -2.092 5.004 32.048 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) 6.411354 0.021168 302.884 < 2e-16 *** d1 0.736294 0.011278 65.285 < 2e-16 *** d2 -0.120134 0.009868 -12.174 < 2e-16 *** d3 0.199693 0.009609 20.783 < 2e-16 *** d4 0.004421 0.007941 0.557 0.578 d5 -0.060423 0.009406 -6.424 1.33e-10 *** d6 0.223852 0.010059 22.253 < 2e-16 *** d7 0.049067 0.008805 5.572 2.51e-08 *** d8 -0.016761 0.008551 -1.960 0.050 * d9 0.247857 0.011252 22.027 < 2e-16 *** d10 -0.188730 0.009733 -19.391 < 2e-16 *** d11 -0.079104 0.009869 -8.016 1.10e-15 *** d12 -0.087549 0.008347 -10.489 < 2e-16 *** --- Signif. codes: 0 â***â 0.001 â**â 0.01 â*â 0.05 â.â 0.1 â â 1 (Dispersion parameter for poisson family taken to be 1) Null deviance: 21668.7 on 69 degrees of freedom Residual deviance: 8720.3 on 57 degrees of freedom AIC: 9350.6 Number of Fisher Scoring iterations: 5
ä¸å¿ãstep()ã§å¤æ°é¸æãè¡ã£ã¦ã¿ã¾ããä¹
ä¿å
çã®ã³ã¡ã³ãã«å¾ããä¸æ¦ããã§ã¯äº¤äºä½ç¨ã®ãã¨ã¯å¿ãããã¨ã«ãã¾ãã
> dat1.glm2<-step(dat1.glm) Start: AIC=9350.63 click ~ d1 + d2 + d3 + d4 + d5 + d6 + d7 + d8 + d9 + d10 + d11 + d12 Df Deviance AIC - d4 1 8720.6 9348.9 <none> 8720.3 9350.6 - d8 1 8724.1 9352.5 - d7 1 8751.6 9379.9 - d5 1 8761.3 9389.6 - d11 1 8784.1 9412.5 - d12 1 8831.1 9459.5 - d2 1 8868.7 9497.0 - d10 1 9093.1 9721.5 - d3 1 9162.3 9790.6 - d6 1 9206.5 9834.8 - d9 1 9206.6 9835.0 - d1 1 13026.4 13654.7 Step: AIC=9348.94 click ~ d1 + d2 + d3 + d5 + d6 + d7 + d8 + d9 + d10 + d11 + d12 Df Deviance AIC <none> 8720.6 9348.9 - d8 1 8724.5 9350.8 - d7 1 8752.5 9378.9 - d5 1 8762.0 9388.3 - d11 1 8784.2 9410.5 - d12 1 8831.8 9458.2 - d2 1 8869.2 9495.5 - d10 1 9101.1 9727.5 - d3 1 9172.8 9799.2 - d9 1 9206.9 9833.2 - d6 1 9209.6 9835.9 - d1 1 13027.5 13653.9 > summary(dat1.glm2) Call: glm(formula = click ~ d1 + d2 + d3 + d5 + d6 + d7 + d8 + d9 + d10 + d11 + d12, family = poisson, data = dat1) Deviance Residuals: Min 1Q Median 3Q Max -23.118 -9.118 -2.065 4.919 32.058 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) 6.413185 0.020914 306.643 < 2e-16 *** d1 0.736299 0.011281 65.271 < 2e-16 *** d2 -0.120217 0.009867 -12.183 < 2e-16 *** d3 0.200392 0.009528 21.032 < 2e-16 *** d5 -0.059299 0.009187 -6.455 1.08e-10 *** d6 0.224234 0.010039 22.336 < 2e-16 *** d7 0.049432 0.008782 5.629 1.81e-08 *** d8 -0.016869 0.008549 -1.973 0.0485 * d9 0.247561 0.011240 22.026 < 2e-16 *** d10 -0.189420 0.009656 -19.616 < 2e-16 *** d11 -0.078933 0.009865 -8.002 1.23e-15 *** d12 -0.087681 0.008344 -10.509 < 2e-16 *** --- Signif. codes: 0 â***â 0.001 â**â 0.01 â*â 0.05 â.â 0.1 â â 1 (Dispersion parameter for poisson family taken to be 1) Null deviance: 21668.7 on 69 degrees of freedom Residual deviance: 8720.6 on 58 degrees of freedom AIC: 9348.9 Number of Fisher Scoring iterations: 5
åºæ¬çã«ã¯åå帰ä¿æ°ã§ããEstimateããã©ã¹ã®å¤ã§ããã°ã¯ãªãã¯æ°å¢ã«è²¢ç®ããéã«ãã¤ãã¹ã®å¤ã§ããã°ã¯ãªãã¯æ°æ¸ã«ã¤ãªãã£ã¦ãã¾ãã¾ããã¾ãEstimateã®å¤ã®å¤§ããï¼çµ¶å¯¾å¤ï¼ãã¯ãªãã¯æ°ã¸ã®å½±é¿ã®å¤§å°ã表ãã¾ãã
ã¨ãããã¨ãè¸ã¾ãã¦èªã¿è§£ãã¦ã¿ãã¨ãã©ã®åå帰ä¿æ°ãçµ±è¨çã«ã¯ææã¨ããå¤å®ã«ãªã£ã¦ããç¹ãæ°ã«ãªããã®ã®ãã¨ããããd1ã®ãã¶ã¤ã³è¦ç´ ãã¯ãªãã¯æ°å¢ã«å¤§ããè²¢ç®ãã¦ãããã¨ãåããã¾ããä»ã«ãã¯ãªãã¯å¢ã«è²¢ç®ããããªã®ã¯d9, d6, d3ãéã«ã¯ãªãã¯æ¸ã«ã¤ãªãã£ã¦ãã¾ããããªã®ã¯d10,d2ã¨ãããã¨ãè¦ã¦åãã¾ãã
è² ã®äºé åå¸ã«ããGLMã§ãã£ã¦ã¿ã
ã¨ããã§ããããããã®ãã¼ã¿ã®clickã«ã©ã ã«ã¤ãã¦ä»¥ä¸ã®ããã«è¨ç®ãã¦ã¿ãã¨ããã
> mean(dat1[,13]) [1] 1067.5 > var(dat1[,13]) [1] 309571.6
ã¨ãªãã®ãè¦ãã°åããéããå¹³åå¤ããåæ£ã®æ¹ããã£ã¨å¤§ããããã¢ã½ã³åå¸ããã å½ã¦ã¯ããã®ã¯æ¬æ¥ãªãã°ãã¾ãé©åã§ã¯ãªãããã§ããã¾ããã¢ã½ã³å帰ã¢ãã«ã«ããçµæãè¦ãã¨"Residual deviance" / "degree of freedom"ã®å¤ãé常ã«å¤§ãã*3ãã¢ãã«ã®å½ã¦ã¯ã¾ãã¨ãã¦ã¯è¯ããªãã¨ãããã¨ãåããã¾ããããã§ã¯ãå¤æ°éè¦åº¦ã¯åãããããªæ°ããã¦ãã¢ãã«ã¨ãã¦ã¯ã¡ãã£ã¨ãããããªãæãã§ãã
ããã§ãä¹ ä¿å çã®ã¢ããã¤ã¹ã«å¾ã£ã¦è² ã®äºé åå¸ã«ããGLMãå°å ¥ãã¦ã¿ã¾ããããè² ã®äºé åå¸ã«ã¤ãã¦ã¯群馬大の青木先生による解説ããMatlabヘルプサイトの解説ãåããããããã¨æãã¾ãã
åãåå¼·ä¸ãªã®ã§ç解ä¸è¶³ãªã®ã§ããããã®ã¢ãã«ã§ã¯ãã¢ã½ã³åå¸ã«æ¯ã¹ã¦å¹³åå¤ã¨åæ£ã¨ã®ä¹é¢ã大ãããããªã«ã¦ã³ããã¼ã¿ã¾ã§å«ãã¦åºæ±ã«è¡¨ç¾ã§ããè² ã®äºé åå¸ã使ããããã«æ£ç¢ºã«ã¢ããªã³ã°ãããã¨ãç®æãã¦ãããã¨ã®ãã¨ã§ãããã®å帰å¼ã¯ä»¥ä¸ã®éãã
ãããRã®glm(){stats}ã§ãããã¨ãã¦ããªãã·ã§ã³ããªãã®ã§ãæ°ããglm.nb(){MASS}ã§ãããã¨ã«ãªãã¾ãã
> dat1.glmnb<-glm.nb(click~.,dat1) > summary(dat1.glmnb) Call: glm.nb(formula = click ~ ., data = dat1, init.theta = 6.40747075, link = log) Deviance Residuals: Min 1Q Median 3Q Max -2.4451 -0.7618 -0.1643 0.4616 2.5026 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) 6.48958 0.26216 24.754 < 2e-16 *** d1 0.71567 0.13279 5.389 7.07e-08 *** d2 -0.15583 0.13574 -1.148 0.2510 d3 0.23391 0.11288 2.072 0.0382 * d4 -0.05671 0.10433 -0.544 0.5868 d5 -0.04942 0.12058 -0.410 0.6819 d6 0.15940 0.13581 1.174 0.2405 d7 0.11804 0.10915 1.081 0.2795 d8 -0.08133 0.11470 -0.709 0.4783 d9 0.19946 0.13711 1.455 0.1457 d10 -0.18792 0.12117 -1.551 0.1209 d11 -0.08563 0.12386 -0.691 0.4893 d12 -0.07408 0.10978 -0.675 0.4998 --- Signif. codes: 0 â***â 0.001 â**â 0.01 â*â 0.05 â.â 0.1 â â 1 (Dispersion parameter for Negative Binomial(6.4075) family taken to be 1) Null deviance: 158.029 on 69 degrees of freedom Residual deviance: 71.921 on 57 degrees of freedom AIC: 1052.4 Number of Fisher Scoring iterations: 1 Theta: 6.41 Std. Err.: 1.07 2 x log-likelihood: -1024.355
ããã§"Residual deviance" / "degree of freedom"ã®å¤ã1.26ãããã¨å¦¥å½ãªæ°åã«åã¾ããã¢ãã«ã¨ãã¦ã®å½ã¦ã¯ã¾ããæªããªãã¨ããçµæã«ãªãã¾ããã
åå帰ä¿æ°ã§çµ±è¨çã«ææãªã®ã¯d1ã¨d3ã®ã¿ã§ãï¼ãã¢ã½ã³å帰ã¢ãã«ã®æ¡å¼µã ããå½ããåãªãã§ããï¼ã ããããã¢ã½ã³å帰ã¢ãã«ã®æã¨è¿ãå¤ã«ãªã£ã¦ãã¾ãã
ã©ã³ãã ãã©ã¬ã¹ãå帰ã§ãã£ã¦ã¿ã
å®ã¯ããã®æã®å帰ã¢ãã«ãã©ã³ãã ãã©ã¬ã¹ãã§ããç¨åº¦è¨ç®ãããã¨ãã§ãã¾ãããã ãã©ã³ãã ãã©ã¬ã¹ãã§ã¯åå帰ä¿æ°ã®ã符å·ãã¾ã§ã¯åºãã¾ããã®ã§ãåºæ¬çã«ã¯å¤æ°éè¦åº¦ãè¦ãã ãã¨ãããã¨ã«ãªãã¾ãã
> dat1.rf<-randomForest(click~.,dat1,ntree=2000) > print(dat1.rf) Call: randomForest(formula = click ~ ., data = dat1, ntree = 2000) Type of random forest: regression # åé¡åé¡ã§ã¯ãªãå帰åé¡ Number of trees: 2000 No. of variables tried at each split: 4 Mean of squared residuals: 181438.4 % Var explained: 40.54 > importance(dat1.rf) IncNodePurity d1 5556729.2 d2 1951166.3 d3 979182.8 d4 679803.4 d5 591530.5 d6 891261.9 d7 678428.5 d8 1087023.2 d9 3094248.6 d10 661204.0 d11 557049.7 d12 486588.2
ã¨ãããã¨ã§é ä½ä»ãããã¦ããã¨ãd1 -> d9 -> d2 -> d8 -> d3 -> d6 -> â¦ã®é ã«å½±é¿ã大ããã¨ãããã¨ãåããã¾ããããããä¸ã®ãã¢ã½ã³åå¸ / è² ã®äºé
åå¸ã«ããGLMã§ç®åºããåå帰ä¿æ°ã®ã符å·ï¼ãã©ã¹orãã¤ãã¹ï¼ãã¨è¦æ¯ã¹ã¦ã¿ãã¨ãå½±é¿ã®å¤§ããã§ä¸ä½ã«æ¥ã6å¤æ°ã®ãã¡ãd2ã¨d8ããã¤ãã¹ï¼ï¼ã¯ãªãã¯æ°æ¸ã«ã¤ãªããï¼ã¨ãããã¨ã«ãªãã®ã§ãçµæ§å«ãªãã¼ã¿ã ã¨æãã¾ãã
åèè³æãªã©
ä»åã®ãã¼ãã§è¨äºãæ¸ãã«å½ãã£ã¦ã以ä¸ã®ãã¼ã¸ãåèã«ãã¾ããã
æ¬å½ã¯å¤é
ãã¸ããã¢ãã«ã¨ãã¾ã§è¡ãããã£ããã§ãããä»åã¯ãã®è¾ºã§ãããããããä»åä½ã£ããµã³ãã«ãã¼ã¿ã¯ãã«ãã³ã¨ãä½ãä¸åèæ
®ãã¦ãªãã£ãããã¦å³å¯ãªãã¨ãè¨ããã¨ããã¨é¢åãªã®ã§ï¼ç¬ï¼ã
ãããããä»åãåã«ã¨ã£ã¦ã¯åå¼·ä¸ã®ãã¼ãã§ããééã£ã¦ããç¹ã»ãããµããªç¹ãªã©ããã°ããã²ãã²ãææä¸ããã
追è¨ï¼SVRã§ãã£ã¦ã¿ãï¼äºæ¸¬ãã¦ã¿ãï¼
ããã¾ã§ãã£ã¦ããã¦SVRï¼ãµãã¼ããã¯ã¿ã¼å帰ï¼ããªãã®ã¯ä½ã ããªãã¨æã£ãã®ã§ã{e1071}ããã±ã¼ã¸ã®SVM*4ã§ãã£ã¦ã¿ã¾ããã
> library("e1071", lib.loc="C:/Program Files/R/R-3.0.1/library") Loading required package: class > dat1.svm<-svm(click~.,dat1) > print(dat1.svm) Call: svm(formula = click ~ ., data = dat1) Parameters: SVM-Type: eps-regression SVM-Kernel: radial cost: 1 gamma: 0.08333333 epsilon: 0.1 Number of Support Vectors: 66
ããã ãã ã¨ã©ãã¢ãã«ãéãã®ãããåãããªãã¨è¨ãããããªã®ã§predict()ã§äºæ¸¬ãããã¨æã£ããã§ãããæè§ãªã®ã§ä»ã®å帰ã¢ãã«ãä½µãã¦äºæ¸¬ãã¦ã¿ããã¨ã«ãã¾ãããã¡ãªã¿ã«äºæ¸¬ã®ããã«å
¥åå¤ã¨ãã¦é¸ãã ã®ã¯ãd1-d12ã¾ã§å
¨ã¦1ã§è¡¨ç¾ããããå
¨é¨çãããã¯ãã«xinã§ãã
> predict(dat1.glm,newdata=xin,type="response") 1 1509.956 > predict(dat1.glmnb,newdata=xin,type="response") 1 1373.509 > predict(dat1.rf,xin) 1 1424.997 > predict(dat1.svm,xin) 1 1287.155
ãããçµæ§ã¢ãã«ãã¨ã«äºæ¸¬çµæãéãã¾ãããã¡ãªã¿ã«d1-d12ã¾ã§å
¨ã¦0ã§è¡¨ç¾ããããå
¨é¨ãªãããã¯ãã«xin0ãå
¥åãã¦äºæ¸¬ããã¨ã
> predict(dat1.glm,newdata=xin0,type="response") 1 608.7171 > predict(dat1.glmnb,newdata=xin0,type="response") 1 658.246 > predict(dat1.rf,xin0) 1 865.2397 > predict(dat1.svm,xin0) 1 805.4882
ããã¾ãçµæ§éãã¾ãããæå¾ã«ããã¢ã½ã³å帰ã¢ãã«ã§å¾ãããåå帰ä¿æ°ã®ã符å·ãã«åããã¦1 / 0ãæ¯ãåãããå¤åã¯ãªãã¯æ°ãæ大å¤ã«ãªãã¯ãããã¯ãã«xinfãå
¥åãã¦äºæ¸¬ããã¨ã
> predict(dat1.glm,newdata=xinf,type="response") 1 2624.216 > predict(dat1.glmnb,newdata=xinf,type="response") 1 2589.834 > predict(dat1.rf,xinf) 1 1679.118 > predict(dat1.svm,xinf) 1 1692.253
çãã¼ã¿ã®æ大å¤ã2325ãªãã§ããããã¢ã½ã³å帰ã¢ãã«ã¨è² ã®äºé
åå¸ã¢ãã«ã§ã¯ãããä¸åãäºæ¸¬å¤ã«ãªã£ã¦ãã¾ããä¸æ¹ã§ãã©ã³ãã ãã©ã¬ã¹ãã¨SVRã«ããäºæ¸¬å¤ã¯ãããä¸åã£ã¦ã¾ãããã¼ã¨ãããä½ã§ã ã£ãã£ããããåå¼·ãã¨ãã¾ãï¼æ±ï¼ãå¤åãä»åã®ãµã³ãã«ãã¼ã¿ã®ãã¡äºæ¸¬ã«å¹ãå¤æ°ãå°ãªãã¦ãã©ã³ãã ãã©ã¬ã¹ãã§ãSVRã§ãå
¨ä½ã®åæ£ã«å¯¾ãã説æåãå¼±ãï¼å®éã©ã³ãã ãã©ã¬ã¹ãã§ã40%ããããã説æã§ãã¦ãªãï¼ãããªãã ã¨æãã¾ããããã
*1:ä»åãé©å½ã«åãrnorm()ãçµã¿åããã¦ä½ã£ãã·ãã¥ã¬ã¼ã·ã§ã³ãã¼ã¿ã§ã
*2:ãææãé ããã®ã§è¨æ£ãã¾ããã http://b.hatena.ne.jp/uncorrelated/20130919#bookmark-161101866 ããã³ http://d.hatena.ne.jp/uncorrelated/20130919/1379599682
*3:ãã®æ¯çã1åå¾ã§ãªãã¨ã¢ãã«ã¨ãã¦ã¯å¦¥å½ã§ã¯ãªãããã
*4:ã¤ã¾ããã®æåãªãã«ããªã³ã¬ã«ããã±ã¼ã¸LIBSVMã®Rç