为 model 选择实现 LOOCV（不使用插入符号包）

Question

I have this dataset where I'm trying to find the best model using R

Dataset:

structure(list(V1 = c(1.43359910241166, 0.411971077467806, 0.236361845246534, 
-0.289263426819727, -1.23202861459847, -0.738796384986188, 0.200420172968439, 
1.55763841132305, 0.306848974278087, -1.06336757529454, 0.208462982445177, 
-0.161933544137143, -0.529226737265933, 1.06311471300635, 0.154281146875831, 
0.609577869238014, -0.13720552696616, 0.920650581183744, 1.18282854178987, 
-0.792945405521446, -0.609722647650392, -0.21688852962299, -3.06426186175807, 
0.5498848363865), V2 = c(-0.322161064448354, -0.203202315321523, 
-1.37681357972322, -2.09183896169083, -1.73416522569493, -0.167163678879473, 
-0.496644140621754, -0.378640254832213, 1.71897857982319, 0.987886990249993, 
-0.464176577243306, 0.313599912560739, 0.279305189424942, 0.621879051693468, 
-1.35413705469938, -0.904307866112488, 0.563960402008738, 0.942178870082166, 
1.05504527675313, 1.72684177309, 0.487583880103759, 0.366982237506534, 
0.341207392409481, 0.0878011635613361), V3 = c(-3.06259779143185, 
0.113156471002083, -0.596111339640452, -0.0549465535711572, -0.941898864240695, 
-0.653015082018507, -0.169956676284042, -0.35411953808696, 0.713862293279259, 
1.20019049753438, 0.295042002436139, -0.248609439893179, 1.9312167684667, 
0.674670687298312, 0.224140747830105, -0.59349261052001, 0.0558808922143246, 
0.749007982254512, 1.04584894162381, 0.280651184742914, -0.313568542992107, 
-1.54267082673779, 0.397265080266878, 0.850053716467332), V4 = c(-2.72697312636474, 
0.851743869193346, -0.0599187094978506, 0.341978048955579, -0.484015693411596, 
-0.131475393689722, -0.021866557862478, -1.8191792655517, 1.74883985589495, 
-0.446343374015597, 0.0107633789594956, 0.55528371030783, 0.31132242799237, 
0.0710046563366782, 0.701388784100771, 1.56870481640847, -0.841113890934613, 
-0.881987858407386, 1.37693978208629, -0.488560120797117, 0.366895195216852, 
0.0627972059134885, -0.655416452787133, 0.589188711953821), V5 = c(-1.79836984688233, 
0.50295466271361, -1.17227869532777, 0.661412408202374, 0.853890060320874, 
0.349725611664228, -0.308069063888987, -0.433246608902138, -0.178767449882736, 
1.34125510863996, 0.206474174580616, -0.657831069822233, 0.215632332747088, 
0.573672331330443, -0.202823754124207, 0.609758501891791, 0.222044482387977, 
2.56037433110525, -1.29345283990688, 0.174550400877521, -0.174265216769768, 
0.55419775558349, -0.458225457879011, -2.14861215865916), V6 = c(-0.18026818728965, 
-0.480816154309526, -0.50256960223903, -1.31874854057412, -0.896086924318379, 
-1.79382217103909, -1.60213450587948, -0.481119812364401, 0.377075792056211, 
1.34981730088023, 0.0611706096060544, 0.83874651540465, 0.58899516399665, 
1.24066391945654, -1.08080170411743, 0.597620326597847, -1.21365483260366, 
0.230893469563153, -0.576677068566099, 1.31703258659203, 0.35136844419016, 
0.925208426922233, 1.73348977742475, 0.514617170610343), V7 = c(0.692646184527114, 
1.64958468445801, -0.722861261417701, -0.411292490473929, -1.73926867251488, 
0.479847732965793, 0.224291785874008, -0.650661070391403, -0.20779505689401, 
-0.900990363217965, 0.712570690351891, 0.0291624484927884, 0.613871305452367, 
-0.901767959624604, -0.184130922600279, 2.60941994159236, 0.0144701586285878, 
1.00941096184201, -1.07148389565784, -0.439790917550134, -0.786567592396622, 
0.926243735906836, -1.39392614240757, 0.449016715055174), V8 = c(-0.218730876718155, 
0.279536175230915, -0.860839531512879, 1.62382620633742, -0.656202640703168, 
-3.05801703213563, 0.243884147081474, 0.926579301241956, 0.58184138659717, 
-0.0814078168437784, -0.0668035158044736, 0.00153834639170001, 
0.806767895958209, 0.834326360087515, -0.0790896439523125, 0.07028192584928, 
-0.619273530317688, 1.07556660504801, -1.13473924521572, 0.668145147063421, 
0.758090513962191, 0.456430947715887, -1.73160959029873, 0.179898464937389
), V9 = c(2.56974590352874, -0.263155790779132, 0.646658371629822, 
-0.752843366448987, 0.200047856906594, 0.659371008337854, 1.24620285734473, 
0.94634794321528, -1.3304334794271, 1.33090401796431, -0.819840444239054, 
0.272969704571894, -0.486961950780986, 0.169639870524667, -0.451658048721127, 
-1.04537018765646, -1.16107891054576, -1.20995090654021, -0.839823653138378, 
0.62253221198192, 0.622634591405887, -0.547608828939565, 0.786557248787584, 
-1.16488601898254), V10 = c(-2.26412916115509, 0.67348993363598, 
-0.342027192999345, 0.249815496496033, 0.30352488488975, -0.744451635640458, 
1.58487417838063, -1.01570448604582, -0.541105970352036, 1.13647671257197, 
-0.54886598448313, -0.962789161396563, -0.538065955333129, 0.0781727823942247, 
0.0970193660300894, 1.18927210039089, -0.6957686086705, -0.386785336508124, 
-0.35257548033064, 2.31937096293864, -0.549132531058022, -0.0974568592721698, 
1.43853645612397, -0.0316945106071529), V11 = c(-1.86095070927053, 
0.573330283491408, -1.03183858717977, -1.83745190916475, -0.077180684913356, 
-0.94533768863225, -0.641638632478328, 0.154349543995556, 1.89664953662371, 
1.3494700201932, 1.04343452008192, 1.03948878970461, 0.394740150081754, 
1.24869842481551, 0.33270007318232, 0.373677276693529, 0.670774298645023, 
-0.0191045174843475, 0.0901593335518681, -0.813757209813031, 
-0.527741614949631, -1.55637393322463, -0.0817683516977811, 0.225671587747989
), V12 = c(0.235155165117673, 0.0334071835637513, 0.141983465568844, 
0.441692874434554, 0.0707526888389656, 0.332161357520943, 0.0735800395703528, 
-0.281305763416249, 0.16538364649173, -1.15487983901285, 1.56899928098857, 
-0.567750194144175, 0.541218236160627, 1.48159680904495, -0.568523352759803, 
-0.0545712227404042, -2.93340050534491, 0.662421496450859, 1.11729205722267, 
-0.581175560009803, 0.792548304722282, 0.955149345977461, -0.821090667653583, 
-1.65064484659245), V13 = c(-1.97412125867671, 0.44572205242864, 
-0.274712915255066, -1.44692140049933, -1.18035700830368, -0.260286573948736, 
-0.95815595797825, -0.242760674716397, 0.477953228907608, 0.992878959448502, 
0.48518262700317, -0.882424015844636, 2.03856721097186, 0.782640940939034, 
0.00789969362112054, -0.295894328060507, 1.27922468162261, 0.51472928905797, 
0.0447383908218823, 0.165638463053774, -0.263332324321804, -1.15204704327981, 
-0.258342890933598, 1.95418085394235), V14 = c(-0.181993529177506, 
1.39403983793056, -0.152733307069606, -1.52421030170283, -0.924924418962197, 
-0.364387222675804, 1.10283509955152, 0.0727783277608945, -1.77522562543095, 
1.08664918075833, -1.04803884297856, -0.940631906527986, 1.12617755875177, 
1.21705368328955, -0.279102677856877, 0.343713803473868, 1.26542530994074, 
-0.774396836280874, 0.417125600747737, 1.49096714826284, 0.284166748008431, 
-1.53295609357739, 0.105608954195959, -0.407940490431605), V15 = c(-1.46474265513464, 
1.19486941463858, 0.244933071673175, -0.459011700723317, 0.241718140420906, 
0.282959623977014, 0.00585677416957126, -2.03400384857495, 0.537918956631718, 
-1.04030075327707, -0.557219563096931, -0.252427064540924, 0.547956268292219, 
-0.526158422645334, 0.251554548033225, -0.745912076395139, -0.0351666299711204, 
1.15204026955591, 0.842246979246097, 1.52268303136091, -1.90156582122334, 
-0.142035061237368, 0.385224459566802, 1.94858205925399), V16 = c(0.828548104520814, 
0.713189024971904, 0.774573684318552, -0.425568343697551, 0.259608074896051, 
-1.22029633555545, -0.344755278537263, 0.973749897026122, -0.474553098183039, 
0.0257155566445092, -0.476287023663646, 0.974669054546108, -1.77164686907544, 
1.56028342699847, 1.24959541751606, -0.574201649578301, 1.2099741843225, 
-0.0750690376790856, -0.0110241372862062, -0.984530244128971, 
-2.52086075001167, 0.0287667805602271, 0.731343831738835, -0.451224270663529
), V17 = c(-0.681074029216176, -0.0390433509889875, 0.0328512523391066, 
1.12428796011696, 0.176765286103444, -0.222850967042728, 0.988520019729737, 
2.09179105565111, 0.116819106946508, 0.51447781508645, 1.87648378755979, 
-1.08036997332246, -0.418517756914466, 0.291253915397003, -0.355756145391065, 
0.874359244531183, -2.35192438381252, -0.200559130397419, -1.29305021151605, 
-0.216777649470054, -1.43207151780606, -0.392317470556723, 0.447601162558867, 
0.149101980414553), V18 = c(-1.96475300593026, 0.422711683040055, 
-1.12996029903421, -2.33587910613298, 0.179352498545959, -0.600058127770143, 
-1.35077156778998, -0.727365308346169, 1.43052873254504, 1.07048786910024, 
1.15649152054786, 0.702163956193049, 0.599458156020645, 0.489172517239038, 
0.957116387643539, 0.335186798948586, -0.598777825023964, 0.10012893280699, 
0.0822063408722808, 0.393896776121708, 0.968441995451939, -0.625513747288306, 
-0.437871585012806, 0.883606407251895), V19 = c(0.203243289070699, 
0.206783154660488, 0.0730205054389099, 0.151752499129077, 0.339065300597841, 
0.198750153846351, 0.246574181097875, 0.219716854159337, 0.112571755773366, 
0.108437458425644, 0.159923853880819, 0.198217376539615, 1.27794667790059, 
0.0628191359027579, -0.023668700184257, 0.0103470645871769, -4.55192891533295, 
0.0932248108210876, 0.0372915017676821, 0.103290843005291, 0.1485089149749, 
0.167015138770557, 0.258108289841612, 0.198988855325523), V20 = c(-0.6885610185506, 
0.215106818871655, -1.26229703607397, -1.15415874394993, -0.770942786330788, 
-1.07811513531511, -1.34581518035362, 0.296281823344214, -0.525449013409778, 
1.52659228597052, 1.66011376586839, 0.204981756466606, 2.25710524990656, 
0.850893107617607, 0.181598239123184, 0.0790398588000734, -0.0665218787774753, 
0.411298611581292, 0.0839458342094344, -0.122405563089466, -1.6897393933796, 
1.24061257187769, -0.157685318761091, -0.145878855645788), outcome_var = c(-3, 
4, 1, -1, -1, -3, -1, -3, 3, 2, -2, -3, 1, 0, 0, 0, 3, 0, 2, 
2, 1, -3, 1, 0)), class = "data.frame", row.names = c(NA, -24L
)) 

Code:

train.control <- trainControl(method = "LOOCV")

step.model <- train(outcome_var ~., data = total,
                method = "leapSeq", 
                tuneGrid = data.frame(nvmax = 1:5),
                trControl = train.control
)

step.model$results

summary(step.model$finalModel)

Result:

20 Variables  (and intercept)
Forced in Forced out
V1      FALSE      FALSE
V2      FALSE      FALSE
V3      FALSE      FALSE
V4      FALSE      FALSE
V5      FALSE      FALSE
V6      FALSE      FALSE
V7      FALSE      FALSE
V8      FALSE      FALSE
V9      FALSE      FALSE
V10     FALSE      FALSE
V11     FALSE      FALSE
V12     FALSE      FALSE
V13     FALSE      FALSE
V14     FALSE      FALSE
V15     FALSE      FALSE
V16     FALSE      FALSE
V17     FALSE      FALSE
V18     FALSE      FALSE
V19     FALSE      FALSE
V20     FALSE      FALSE
1 subsets of each size up to 3
Selection Algorithm: 'sequential replacement'
         V1  V2  V3  V4  V5  V6  V7  V8  V9  V10 V11 V12 V13 V14 V15 V16 V17 V18 V19 V20
1  ( 1 ) " " " " "*" " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "
2  ( 1 ) " " " " "*" " " " " " " " " " " " " " " " " " " " " " " " " " " "*" " " " " " "
3  ( 1 ) " " " " " " " " " " " " " " " " " " "*" " " " " "*" " " " " " " "*" " " " " " "

Which gives me the output I'm looking for, but now I'm trying to make my own LOOCV function and not use the caret package for it. And I'm not getting the same results,

loocv = function(fit) {
  n = length(fit$residuals)
  yvar = fit$model[, 1]
  index = 1:n
  e = rep(NA, n)
  for (i in index) {
    refit = update(fit, subset = index != i)
    pred = predict(refit, dplyr::slice(fit$model, i))
    e[i] = yvar[i] - pred
  }
  return(mean(e^2))
}

How can I use LOOCV without using the caret package and find the best fitting model?

Answer 1

For cross-validation like LOOCV , the model should be constructed from scratch for each testing fold. By trial and error, I believe caret uses leaps::regsubsets for stepwise model selection.

library(leaps)

nvmax = 3 #number of max variables
pred = rep(NA, nrow(total))
for (i in seq(nrow(total))) #LOOCV
  {#train a new model
   tem = regsubsets(x=total[-i,1:20], 
                    y=total[-i,21], 
                    nvmax=nvmax, 
                    method="seqrep")
  coef(tem, nvmax) #best coef chosen
  fit = lm(outcome_var ~ ., 
           data = total[-i,
                  c(which(summary(tem)$which[nvmax,-1]), 
                  21)])

  #predict the hold-out data
  pred[i] = predict(fit, newdata=total[i,])
  }

RMSE(pred, total[,'outcome_var'])
#1.945036

MAE(pred, total[,'outcome_var'])
#1.442353

Results from caret:

step.model$results
# nvmax     RMSE    Rsquared      MAE
#     3 1.945036 0.238655497 1.442353