Issue when predicting probabilities with multinom and one observation

[YvesCR]

When predicting probability from a multinom model, if there is only one observation in the new data set, the predict function of the train class is flashing an error.

Reproductible exemple:

pacman::p_load(nnet, caret)

training <- iris[-1, ]
testing <- iris[1, ]
testing2 <- iris[1:2, ]

fitControl <- trainControl(## 4-fold CV
  method = "cv",
  number = 4)

set.seed(1789)
multinomFit1 <- train(Species ~ ., data = training, 
                 method = "multinom", 
                 trControl = fitControl)
multinomFit1

# ok
predict(multinomFit1, newdata = training, type = "prob")
# ok
predict(multinomFit1, newdata = testing2, type = "prob")
# not ok
predict(multinomFit1, newdata = testing, type = "prob")

Error:

Error in dimnames(x) <- dn : 
  length of 'dimnames' [2] not equal to array extent

I guess it comes from the fact that output of predict.nnet in case of multiple observations is a matrix whereas in the case of one observation it is a named vector.

Session Info:

R version 3.4.1 (2017-06-30)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United Kingdom.1252
[2] LC_CTYPE=English_United Kingdom.1252
[3] LC_MONETARY=English_United Kingdom.1252
[4] LC_NUMERIC=C
[5] LC_TIME=English_United Kingdom.1252

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

other attached packages:
[1] caret_6.0-76 ggplot2_2.2.1 lattice_0.20-35 nnet_7.3-12
[5] MASS_7.3-47

loaded via a namespace (and not attached):
[1] Rcpp_0.12.11 magrittr_1.5 splines_3.4.1
[4] munsell_0.4.3 colorspace_1.3-2 rlang_0.1.2
[7] foreach_1.4.3 minqa_1.2.4 stringr_1.2.0
[10] car_2.1-5 plyr_1.8.4 tools_3.4.1
[13] parallel_3.4.1 pbkrtest_0.4-7 grid_3.4.1
[16] gtable_0.2.0 nlme_3.1-131 mgcv_1.8-17
[19] pacman_0.4.6 quantreg_5.33 e1071_1.6-8
[22] class_7.3-14 MatrixModels_0.4-1 iterators_1.0.8
[25] lme4_1.1-13 lazyeval_0.2.0 tibble_1.3.3
[28] Matrix_1.2-10 nloptr_1.0.4 reshape2_1.4.2
[31] ModelMetrics_1.1.0 codetools_0.2-15 stringi_1.1.5
[34] compiler_3.4.1 scales_0.5.0 stats4_3.4.1
[37] SparseM_1.77

[topepo]

That should do it but please test on your data. If you don't want to install from github, you can source the file in models/files/multinom.R and use method = modelInfo.