Getting a strange message from H2O ( h2o_3.26.0.2 ) when predicting using a MOJO file:
Detected 14 unused columns in the input data set: {X8,X9,X10,X12,X1,X11,X2,X14,X3,X13,X4,X5,X6,X7}
I know that it is not a missing variable issue, as then H2O outputs:
There were 1 missing columns found in the input data set: {X1}
To reproduce the warning message I have created a small example with X15 being my target variable:
suppressPackageStartupMessages(library(tidyverse))
suppressPackageStartupMessages(library(h2o))
# read in data ------------------------------------------------------------
data_set <- read_csv("https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data", col_names = FALSE) %>%
mutate_if(is.character, factor)
set.seed(3456)
trainIndex <- caret::createDataPartition(data_set$X15, p = .8,
list = FALSE,
times = 1)
train_strat <- data_set[ trainIndex,]
test_strat <- data_set[-trainIndex,]
# start h2o ---------------------------------------------------------------
h2o.init(startH2O = T, max_mem_size = '400G')
#> Connection successful!
#>
#> R is connected to the H2O cluster:
#> H2O cluster uptime: 25 minutes 3 seconds
#> H2O cluster timezone: Etc/UTC
#> H2O data parsing timezone: UTC
#> H2O cluster version: 3.26.0.2
#> H2O cluster version age: 6 months and 29 days !!!
#> H2O cluster name: H2O_started_from_R_rstudio_ltd590
#> H2O cluster total nodes: 1
#> H2O cluster total memory: 353.12 GB
#> H2O cluster total cores: 64
#> H2O cluster allowed cores: 64
#> H2O cluster healthy: TRUE
#> H2O Connection ip: localhost
#> H2O Connection port: 54321
#> H2O Connection proxy: NA
#> H2O Internal Security: FALSE
#> H2O API Extensions: Amazon S3, XGBoost, Algos, AutoML, Core V3, Core V4
#> R Version: R version 3.5.1 (2018-07-02)
#> Warning in h2o.clusterInfo():
#> Your H2O cluster version is too old (6 months and 29 days)!
#> Please download and install the latest version from http://h2o.ai/download/
# h2o.shutdown()
train_strat_h2o <- as.h2o(train_strat)
test_strat_h2o <- as.h2o(test_strat)
Y <- 'X15'
X <- setdiff(names(train_strat), Y)
# train and predict --------------------------------------------------------------
rf_h2o <- h2o.randomForest(
training_frame = train_strat_h2o,
x = X,
y = Y,
nfolds = 0,
model_id = "big_rf",
ntrees = 25,
max_depth = 55,
stopping_rounds = 5,
stopping_tolerance = 1e-3,
score_each_iteration = T,
seed = 123
)
After the model has been trained and is sitting in the environment, I do a prediction and get the following results :
predict(rf_h2o, test_strat_h2o)
#> predict <=50K >50K
#> 1 <=50K 1.0000000 0.0000000
#> 2 >50K 0.4980000 0.5020000
#> 3 >50K 0.2465993 0.7534007
#> 4 >50K 0.1200000 0.8800000
#> 5 <=50K 0.8040000 0.1960000
#> 6 <=50K 0.8628571 0.1371429
#>
#> [6512 rows x 3 columns]
Now I move on to put the model into production by downloading the model:
h2o.download_mojo(rf_h2o, path = "output/", get_genmodel_jar = T)
And finally we can now use the MOJO file to predict. HERE is where I get the odd message, Detected 14 unused columns in the input data set, although the predictions seem the same?
head(
h2o.mojo_predict_df(
test_strat[, -15],
mojo_zip_path = "output/big_rf.zip",
genmodel_jar_path = "output/h2o-genmodel.jar"
)
)
#> Detected 14 unused columns in the input data set: {X8,X9,X10,X12,X1,X11,X2,X14,X3,X13,X4,X5,X6,X7}
#> predict X..50K X.50K
#> 1 <=50K 1.0000000 0.0000000
#> 2 >50K 0.4980000 0.5020000
#> 3 >50K 0.2465993 0.7534007
#> 4 >50K 0.1200000 0.8800000
#> 5 <=50K 0.8040000 0.1960000
#> 6 <=50K 0.8628571 0.1371429
Created on 2020-02-25 by the reprex package (v0.2.1)
Is this something I should be worried about?
