H2O随机森林挂牌完成

Question

I'm training a Random Forest using h2o and R on a large (~6 million) row dataset and ~50 output levels. Despite the progress bar hitting 100% the console (and the processor!) is still busy and hangs for over an hour (so far!). Definitely not resource limitations, I have 120gb of RAM and a couple of dozen cores.

Hard to give a fully reproducible example given the nature of the issue but there are 35 variables, half of which are factors, I'm running the model training through R with the following options:

rforest <- h2o.randomForest(y = y.var
                          , x = x.vars
                          , training_frame = trainData.h2o
                          , validation_frame = testData.h2o
                          , ntrees = 100
                          , stopping_rounds = 3
                          , seed = 42
                          , model_id = modCode
                          , mtries = -1)

Has anyone encountered a similar issue/has an explanation/knows a workaround, please?

Answer 1

Did you do logarithmic transformation of response variable (ie y ) before running the model? If yes, then are you sure you did not have any y = 1 values BEFORE you log-transformed it? I had a similar issue, and model worked really fast after I removed from data set the rows with y = 1 .