I am able to train a Catboost model with caret (in Rstudio) and it works great.
my_catboost <- caret::train(x, y,
method=catboost.caret,
trControl=fitControl,
tuneGrid = param,
metric = "ROC")
If I use the model to predict on new data in the same session, no issue, it works:
output <- caret::predict.train(my_catboost, newdata=x_testing, type="prob")
However, If I save the model and load it later (or save it, delete "my_catboost" and load), the function predict will crash R and Rstudio without error message and can't find anything in Rstudio log. After the load, I can see the model being created in the global Environment and it seems fine.
I tried the R function save and load, saveRDS and readRDS and both crashed
Thanks !
You have misunderstood my comment. Here is an answer using the inbuilt data set Sonar:
library(caret)
library(catboost)
library(mlbench)
data(Sonar)
create train and test data sets:
set.seed(1)
tr <- createDataPartition(Sonar$Class, p = 0.7, list = FALSE)
trainer <- Sonar[tr,]
tester <- Sonar[-tr,]
train models:
fitControl <- trainControl(method = "cv",
number = 3,
savePredictions = TRUE,
summaryFunction = twoClassSummary,
classProbs = TRUE)
model <- train(x = trainer[,1:60],
y = trainer$Class,
method = catboost.caret,
trControl = fitControl,
tuneLength = 5,
metric = "ROC")
predict using caret:
preds1 <- predict(model, tester, type = "prob")
save the final model:
catboost::catboost.save_model(model$finalModel, "model")
load the saved model:
model2 <- catboost::catboost.load_model("model")
predict using the saved model:
preds2 <- catboost.predict(model2,
catboost.load_pool(tester),
prediction_type = "Probability")
check equality of predictions
all.equal(preds1[,2], preds2)
EDIT: while:
saveRDS(model, "caret.model.rds")
model3 <- readRDS("caret.model.rds")
preds3 <- predict(model3, tester, type = "prob")
results in R session crash
R version 3.5.0 (2018-04-23)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)
Matrix products: default
locale:
[1] LC_COLLATE=English_United States.1252 LC_CTYPE=English_United States.1252 LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C LC_TIME=English_United States.1252
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] mlbench_2.1-1 catboost_0.10.3 caret_6.0-80 ggplot2_2.2.1 lattice_0.20-35 RevoUtils_11.0.0
[7] RevoUtilsMath_11.0.0
loaded via a namespace (and not attached):
[1] httr_1.3.1 magic_1.5-8 ddalpha_1.3.3 tidyr_0.8.1 sfsmisc_1.1-2 jsonlite_1.5
[7] viridisLite_0.3.0 splines_3.5.0 foreach_1.5.0 prodlim_2018.04.18 assertthat_0.2.0 stats4_3.5.0
[13] DRR_0.0.3 yaml_2.1.19 robustbase_0.93-0 ipred_0.9-6 pillar_1.2.3 glue_1.2.0
[19] digest_0.6.15 colorspace_1.3-2 recipes_0.1.2 htmltools_0.3.6 Matrix_1.2-14 plyr_1.8.4
[25] psych_1.8.4 timeDate_3043.102 pkgconfig_2.0.1 CVST_0.2-2 broom_0.4.4 purrr_0.2.4
[31] scales_0.5.0 gower_0.1.2 lava_1.6.1 tibble_1.4.2 withr_2.1.2 nnet_7.3-12
[37] lazyeval_0.2.1 mnormt_1.5-5 survival_2.41-3 magrittr_1.5 nlme_3.1-137 MASS_7.3-49
[43] dimRed_0.1.0 foreign_0.8-70 class_7.3-14 tools_3.5.0 data.table_1.11.4 stringr_1.3.1
[49] plotly_4.7.1 kernlab_0.9-26 munsell_0.4.3 bindrcpp_0.2.2 compiler_3.5.0 RcppRoll_0.2.2
[55] rlang_0.2.0 grid_3.5.0 iterators_1.0.10 htmlwidgets_1.2 geometry_0.3-6 gtable_0.2.0
[61] ModelMetrics_1.1.0 codetools_0.2-15 abind_1.4-5 reshape2_1.4.3 R6_2.2.2 lubridate_1.7.4
[67] dplyr_0.7.5 bindr_0.1.1 stringi_1.1.7 parallel_3.5.0 Rcpp_0.12.17 rpart_4.1-13
[73] DEoptimR_1.0-8 tidyselect_0.2.4