Incorrect definition of glm.nb model

[jpclemens0]

The glm.nb model fails to produce suitable output for the default cross validation cost function when fitting count data because it is missing type = 'response' in the definition of predict. Compare with the glm model.

Minimal, reproducible example:

require('caret')
require('MASS')

cost <- function(data, lev = NULL, model = NULL) {
print(head(data))
defaultSummary(data)
}

train_control <- trainControl(method = 'cv', number = 2, summaryFunction = cost)
tr_out <- train(Days ~ Sex/(Age + Eth*Lrn), data = quine, trControl = train_control, method = 'glm.nb', tuneLength = 1)

Session Info:

>sessionInfo()
R version 3.4.0 (2017-04-21)
Platform: x86_64-apple-darwin15.6.0 (64-bit)
Running under: macOS Sierra 10.12.5

Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRlapack.dylib

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] MASS_7.3-47     caret_6.0-76    ggplot2_2.1.0   lattice_0.20-35

loaded via a namespace (and not attached):
 [1] Rcpp_0.12.11       magrittr_1.5       splines_3.4.0      munsell_0.4.3      colorspace_1.3-2   foreach_1.4.3     
 [7] minqa_1.2.4        stringr_1.2.0      car_2.1-5          plyr_1.8.4         tools_3.4.0        parallel_3.4.0    
[13] nnet_7.3-12        pbkrtest_0.4-7     grid_3.4.0         gtable_0.2.0       nlme_3.1-131       mgcv_1.8-17       
[19] quantreg_5.33      MatrixModels_0.4-1 iterators_1.0.8    lme4_1.1-13        Matrix_1.2-10      nloptr_1.0.4      
[25] reshape2_1.4.2     ModelMetrics_1.1.0 codetools_0.2-15   stringi_1.1.5      compiler_3.4.0     scales_0.4.1      
[31] stats4_3.4.0       SparseM_1.77  

[topepo]

You are correct! The new code gives:

> tr_out <- train(Days ~ Sex/(Age + Eth*Lrn), data = quine, trControl = train_control, method = modelInfo, tuneLength = 1)
    pred obs rowIndex
137    1   0       89
23    43  11       20
142    5  19      127
51    19   6       46
91     8  81       94
103   41   0      110
        pred obs rowIndex
1  17.125190   2        1
4  13.649380   5        4
6  13.649380  13        6
7  13.649380  20        7
12  5.210164   7       12
13  5.210164  14       13
       pred obs rowIndex
2  15.44885  11        2
3  15.44885  14        3
5  10.59268   5        5
8  10.59268  22        8
9  13.85107   6        9
10 13.85107   6       10

(the first set is a random check that train does to make sure that the evaluation function is working correctly)