xgboost
xgboost copied to clipboard
dmlc
Suppressing waring message 'WARNING: amalgamation/../src/learner.cc:438'
Recently, I used xgboost v1.6.0.1 in R. When I use a model trained to in old xgboost version to predict, the log comes out:
WARNING: amalgamation/../src/learner.cc:438:
If you are loading a serialized model (like pickle in Python, RDS in R) generated by
older XGBoost, please export the model by calling `Booster.save_model` from that version
first, then load it back in current version. See:
https://xgboost.readthedocs.io/en/latest/tutorials/saving_model.html
for more details about differences between saving model and serializing.
This warning message could not be suppressed by setting verbose=F or verbose=0
pred <- predict(object, newdata, verbose=0)
After reading introduction in https://xgboost.readthedocs.io/en/latest/tutorials/saving_model.html, I still have no idea how to deal with the annoying log. Any suggestions?
By the way, the code and the model works. It's just the problem about the annoying logs.
I used codes like and it doen't work on suppressing the warning message:
xgb.save(old_bst, 'xgb.model')
new_bst <- xgb.load('xgb.model')
if (file.exists('xgb.model')){
file.remove('xgb.model')
pred <- predict(new_bst, newdata)
}
Any suggestions?
This seems like an issue for detecting old model. Let me take a deeper look tomorrow.
Hi, do you mind if you share your model (possibly in private)? I couldn't reproduce it by using models from 1.0.0.
@trivialfis OK！
Here is the Github repositry of my book, where I use xgboost as a method to solve classification problems.
I used some codes based on the GSClassifier package I developed.
# Package
# Install "devtools" package
if (!requireNamespace("devtools", quietly = TRUE))
install.packages("devtools")
# Install dependencies
if (!requireNamespace("luckyBase", quietly = TRUE))
devtools::install_github("huangwb8/luckyBase")
# Install the "GSClassifier" package
if (!requireNamespace("GSClassifier", quietly = TRUE))
devtools::install_github("huangwb8/GSClassifier")
# Load needed packages
library(GSClassifier)
# Data
testData <- readRDS(system.file("extdata", "testData.rds", package = "GSClassifier"))
M <- readRDS(system.file("extdata", "PAD.train_20220916.rds", package = "GSClassifier"))
# Test data and model
X <- testData$PanSTAD_expr_part
X_bined <- GSClassifier:::trainDataProc_X(X[,1:10], geneSet = M$geneSet)
X_bined_matrix <- X_bined$dat$Xbin
model <- M$ens$Model[[1]]$`2`
X_bined_matrix2 <- X_bined_matrix[,model$genes]
# Prediction
pred <- predict(model$bst, X_bined_matrix2)
Here is the actual output containing the warning message about xgboost version via RMarkdown of RStudio/Posit:
Then, I used model$bst to call an xgboost model:
print(model$bst)
The output is like:
##### xgb.Booster
raw: 60.3 Kb
call:
xgb.train(params = params, data = dtrain, nrounds = nrounds,
watchlist = watchlist, verbose = verbose, print_every_n = print_every_n,
early_stopping_rounds = early_stopping_rounds, maximize = maximize,
save_period = save_period, save_name = save_name, xgb_model = xgb_model,
callbacks = callbacks, max_depth = ..1, eta = ..2, nthread = ..3,
objective = "binary:logistic")
params (as set within xgb.train):
max_depth = "10", eta = "0.5", nthread = "10", objective = "binary:logistic", validate_parameters = "TRUE"
xgb.attributes:
niter
callbacks:
cb.print.evaluation(period = print_every_n)
cb.evaluation.log()
# of features: 352
niter: 15
nfeatures : 352
evaluation_log:
iter train_logloss
1 0.400678
2 0.267733
---
14 0.031730
15 0.029224
Finally, the information about the current R session is :
R version 4.0.3 (2020-10-10)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 18363)
Matrix products: default
locale:
[1] LC_COLLATE=Chinese (Simplified)_China.936
[2] LC_CTYPE=Chinese (Simplified)_China.936
[3] LC_MONETARY=Chinese (Simplified)_China.936
[4] LC_NUMERIC=C
[5] LC_TIME=Chinese (Simplified)_China.936
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] GSClassifier_0.1.25 luckyBase_0.1.0
loaded via a namespace (and not attached):
[1] colorspace_2.0-3 ggsignif_0.6.3 rjson_0.2.21
[4] ellipsis_0.3.2 class_7.3-20 rprojroot_2.0.3
[7] circlize_0.4.15 GlobalOptions_0.1.2 fs_1.5.2
[10] clue_0.3-57 ggpubr_0.4.0 listenv_0.8.0
[13] remotes_2.4.2 prodlim_2019.11.13 fansi_1.0.3
[16] lubridate_1.8.0 codetools_0.2-18 splines_4.0.3
[19] doParallel_1.0.17 cachem_1.0.6 knitr_1.30
[22] pkgload_1.2.4 jsonlite_1.8.0 pROC_1.18.0
[25] caret_6.0-92 broom_1.0.0 cluster_2.1.3
[28] png_0.1-7 compiler_4.0.3 backports_1.4.1
[31] assertthat_0.2.1 Matrix_1.4-1 fastmap_1.1.0
[34] cli_3.3.0 htmltools_0.5.2 prettyunits_1.1.1
[37] tools_4.0.3 gtable_0.3.0 glue_1.6.2
[40] reshape2_1.4.4 dplyr_1.0.9 Rcpp_1.0.8.3
[43] carData_3.0-5 vctrs_0.4.1 nlme_3.1-149
[46] iterators_1.0.14 timeDate_3043.102 gower_1.0.0
[49] xfun_0.33 stringr_1.4.0 globals_0.15.1
[52] ps_1.4.0 testthat_3.1.0 lifecycle_1.0.1
[55] devtools_2.4.3 rstatix_0.7.0 future_1.26.1
[58] MASS_7.3-53 scales_1.2.0 ipred_0.9-12
[61] parallel_4.0.3 RColorBrewer_1.1-3 ComplexHeatmap_2.4.3
[64] yaml_2.3.5 memoise_2.0.1 ggplot2_3.3.6
[67] rpart_4.1.16 stringi_1.7.6 desc_1.4.1
[70] randomForest_4.6-14 foreach_1.5.2 hardhat_1.1.0
[73] pkgbuild_1.3.1 lava_1.6.10 shape_1.4.6
[76] tuneR_1.4.0 rlang_1.0.2 pkgconfig_2.0.3
[79] evaluate_0.15 lattice_0.20-41 purrr_0.3.4
[82] recipes_0.2.0 processx_3.7.0 tidyselect_1.1.2
[85] parallelly_1.32.0 plyr_1.8.7 magrittr_2.0.3
[88] bookdown_0.21 R6_2.5.1 generics_0.1.2
[91] DBI_1.1.3 pillar_1.7.0 withr_2.5.0
[94] abind_1.4-5 survival_3.3-1 nnet_7.3-17
[97] tibble_3.1.7 future.apply_1.9.0 crayon_1.5.1
[100] car_3.1-0 xgboost_1.6.0.1 utf8_1.2.2
[103] rmarkdown_2.14 GetoptLong_1.0.5 usethis_2.1.3
[106] grid_4.0.3 data.table_1.14.2 callr_3.7.0
[109] ModelMetrics_1.2.2.2 digest_0.6.29 tidyr_1.2.0
[112] signal_0.7-7 stats4_4.0.3 munsell_0.5.0
[115] sessioninfo_1.2.2
I do hope it can help for your debugging. Thank you very much!
The model training process is like:
cvRes <- xgb.cv(data = dtrain,
nrounds=params$nrounds,
nthread=params$nthread,
nfold=params$nfold,
max_depth=params$max_depth,
eta=params$eta,
early_stopping_rounds=2,
metrics = list("logloss", "auc"),
objective = "binary:logistic",
verbose = verbose)
bst <- xgboost(data = Xbin,
label = Ybin,
max_depth=params$max_depth,
eta=params$eta,
nrounds = cvRes$best_iteration,
nthread=params$nthread,
objective = "binary:logistic",
verbose = ifelse(verbose,1,0))
Any suggestions?
Hi, I appreciate the detailed information, but could you please try to extract a simpler code snippet that I can copy and run? I tried to install your package GSClassifier and ran into version conflicts after debugging issues with other package installations:
Warning message:
package ‘ComplexHeatmap’ is not available for this version of R
It would be great if there's a simpler way that I can get started.
@trivialfis Thanks for your attention!
First, you can download files testData.rds and PAD.train_20220916.rds here: https://github.com/huangwb8/GSClassifier/tree/master/inst/extdata
Second, you can just use codes like:
testData <- readRDS('E:/RCloud/RFactory/GSClassifier/inst/extdata/testData.rds')
M <- readRDS('E:/RCloud/RFactory/GSClassifier/inst/extdata/PAD.train_20220916.rds')
# Test data and model
X <- testData$PanSTAD_expr_part
X_bined <- GSClassifier:::trainDataProc_X(X[,1:10], geneSet = M$geneSet)
X_bined_matrix <- X_bined$dat$Xbin
model <- M$ens$Model[[1]]$`2`
X_bined_matrix2 <- X_bined_matrix[,model$genes]
# Prediction
pred <- predict(model$bst, X_bined_matrix2)
Replace paths of the readRDS function with your local path of these files before runing the above codes.
I do hope it can help!
I'm not sure why https://github.com/dmlc/xgboost/issues/8248#issuecomment-1248940061 doesn't work on your environment.
I tried both the master branch and the 1.6 release (from cran and build from source) and ran this (your script with an extra for loop):
library(xgboost)
testData <- readRDS('./testData.rds')
M <- readRDS('./PAD.train_20220916.rds')
# Test data and model
X <- testData$PanSTAD_expr_part
X_bined <- GSClassifier:::trainDataProc_X(X[,1:10], geneSet = M$geneSet)
X_bined_matrix <- X_bined$dat$Xbin
model <- M$ens$Model[[1]]$`2`
X_bined_matrix2 <- X_bined_matrix[,model$genes]
print("save")
xgb.save(model$bst, "saved-old")
print("load")
loaded <- xgb.load("saved-old")
## model$bst
# Prediction
print("prediction")
for (i in seq(0, 4)) {
print(i)
pred <- predict(loaded, X_bined_matrix2)
}
and here is the output, the warning is only generated during the call to xgb.save and the prediction is silent.
[1] "save"
[16:50:24] WARNING: amalgamation/../src/learner.cc:1040:
If you are loading a serialized model (like pickle in Python, RDS in R) generated by
older XGBoost, please export the model by calling `Booster.save_model` from that version
first, then load it back in current version. See:
https://xgboost.readthedocs.io/en/latest/tutorials/saving_model.html
for more details about differences between saving model and serializing.
[16:50:24] WARNING: amalgamation/../src/learner.cc:749: Found JSON model saved before XGBoost 1.6, please save the model using current version again. The support for old JSON model will be discontinued in XGBoost 2.3.
[16:50:24] WARNING: amalgamation/../src/learner.cc:438:
If you are loading a serialized model (like pickle in Python, RDS in R) generated by
older XGBoost, please export the model by calling `Booster.save_model` from that version
first, then load it back in current version. See:
https://xgboost.readthedocs.io/en/latest/tutorials/saving_model.html
for more details about differences between saving model and serializing.
[1] TRUE
[1] "load"
[1] "prediction"
[1] 0
[1] 1
[1] 2
[1] 3
[1] 4
Yep, that's the problem!
The code above was just a simple example.
During real model training and subtype calling via GSClassifier, similar process would be done lots of time, which means lots of similar Warning, which are redundant and annoying.
Well, the model actually works. I just want to hide warning messages during the usage of xgboost, if possible. However, neither verbose=F nor verbose=0 could help.
Any suggestions?
You mentioned that you tried to save and load the model using the latest xgboost https://github.com/dmlc/xgboost/issues/8248#issuecomment-1248940061 . Can you replace the old model with the new one?
@trivialfis Thanks for your suggestions! I think it is the best solution. The models should updated to the latest version to avoid warning.