Reputation: 544
I have built a GBM model in R with the below code.
gbm_model_sample <- h2o.gbm(x = c(1:78,80:688), y =79, training_frame = train.h2o, seed = 0xDECAF,ntrees = 1000, max_depth = 4,learn_rate = 0.1,stopping_rounds=50,min_rows = 50,distribution ="bernoulli",ignore_const_col=F,
histogram_type='QuantilesGlobal',sample_rate=0.7,col_sample_rate=0.7,keep_cross_validation_models = T)
The model gets built and i save the Mojo object as :
h2o.download_mojo(gbm_model_sample,get_genmodel_jar = T)
which is saved as "GBM_model_R_1586221409024_1.zip" in my working directory.
Now i use function h2o.mojo_predict_csv
and/or h2o.mojo_predict_df
to predict on test data frame which is where i get the error as below
for h2o.mojo_predict_csv
h2o.mojo_predict_csv('Test_sample_.csv','GBM_model_R_1586221409024_1.zip',genmodel_jar_path = 'h2o-genmodel.jar',verbose = F)
for h2o.mojo_predict_df
h2o.mojo_predict_df(test, 'GBM_model_R_1586221409024_1.zip',verbose = T)
when i use the same test and use the within R h2o.predict
it works completely fine , however the above two codes which had been working fine for me before have started giving the errors as above. My packages loaded are as below. What is causing this error? i haven't manage to find much info on this online.
library(rJava)
require(h2o)
require(readr)
require(dplyr)
require(forcats)
require(ggplot2)
require(scales)
require(caret)
require(stringr)
library(data.table)
require(getPass)
Upvotes: 0
Views: 169
Reputation: 544
As silly as it sounds, there seems to be a bug in h2o which occurs when have a working directory set which has empty spaces in its name. e.g. "c:\test folder\model\"
, if you change this to "c:\test_folder\model\
" or "c:\testfolder\model\"
, then we don't get the above error. H2o has difficult writing the files to those directories with address where there are empty spaces in between.
Upvotes: 2