在公式中包含多个模型列时,Caret :: calibration复制数据

时间:2019-05-22 17:45:47

标签: r r-caret lattice

不确定这是一个错误还是我的理解存在缺陷,但是当我运行以下示例时:

library(caret)
data(mdrr)
mdrrDescr <- mdrrDescr[, -nearZeroVar(mdrrDescr)]
mdrrDescr <- mdrrDescr[, -findCorrelation(cor(mdrrDescr), .5)]
preProc <- preProcess(mdrrDescr, c("center", "scale"))
mdrrDescr <- predict(preProc, mdrrDescr)

inTrain <- createDataPartition(mdrrClass)
trainX <- mdrrDescr[inTrain[[1]], ]
trainY <- mdrrClass[inTrain[[1]]]
testX <- mdrrDescr[-inTrain[[1]], ]
testY <- mdrrClass[-inTrain[[1]]]

library(MASS)

ldaFit <- lda(trainX, trainY)
qdaFit <- qda(trainX, trainY)

testProbs <- data.frame(obs = testY,
                    lda = predict(ldaFit, testX)$posterior[,1],
                    qda = predict(qdaFit, testX)$posterior[,1])

calPlotData <- caret::calibration(obs ~ lda + qda, data = testProbs, cuts = 5)

> calPlotData$data

我得到这个结果:

# out: 
calibModelVar       bin   Percent     Lower    Upper    Count   midpoint
1            lda   [0,0.2]  6.521739  2.430775 13.65621     6       10
2            lda (0.2,0.4] 30.232558 20.789989 41.08301    26       30
3            lda (0.4,0.6] 59.375000 46.367688 71.48530    38       50
4            lda (0.6,0.8] 70.909091 61.481025 79.17690    78       70
5            lda   (0.8,1] 85.227273 79.108431 90.11742   150       90
6            qda   [0,0.2] 28.099174 22.529270 34.21445    68       10
7            qda (0.2,0.4] 40.000000 12.155226 73.76219     4       30
8            qda (0.4,0.6] 33.333333  9.924609 65.11245     4       50
9            qda (0.6,0.8] 80.000000 56.338600 94.26660    16       70
10           qda   (0.8,1] 84.426230 79.256188 88.73729   206       90

但是,当我进行一些调查时,发现这些结果中的数据是重复的。例如,

>table(testProbs$obs == "Active" & testProbs$lda <= 0.2)
# out:
FALSE  TRUE 
 261     3

>table(testProbs$obs == "Active" & testProbs$qda <= 0.2)
# out:
FALSE  TRUE 
 230    34 

这也影响误差估计(表中的上和下)。例如,当我仅使用模型列之一运行Calibration()函数时,

> calPlotData <- caret::calibration(obs ~ lda, data = testProbs, cuts = 5)
> calPlotData$data
# out:
   calibModelVar    bin   Percent     Lower    Upper    Count   midpoint
1           lda   [0,0.2]  6.521739  1.365677 17.89644     3       10
2           lda (0.2,0.4] 30.232558 17.182499 46.12533    13       30
3           lda (0.4,0.6] 59.375000 40.644925 76.30159    19       50
4           lda (0.6,0.8] 70.909091 57.101742 82.37003    39       70
5           lda   (0.8,1] 85.227273 76.063784 91.89296    75       90

1 个答案:

答案 0 :(得分:0)

这是一个错误,已由this PR修复。