MLR and randomForestSRC: Incoherence between computers

Question

Our team run the following code to create a random forest model and train it:

# Define a cross validation strategy
rdesc <- makeResampleDesc("CV", iters = cv_fold, predict = "both")

# Define a (regression) task
task_01 = makeRegrTask(data = data.model, target = "target_actual")

# Make a learner
lrn_rf = makeLearner("regr.randomForestSRC", predict.type = "response",
                     fix.factors.prediction = TRUE,
                     par.vals = list(nodesize = 50, mtry = 36, ntree = 500))

set.seed(7)
model_rf = mlr::resample(lrn_rf, task_01, rdesc, models = TRUE, 
                         extract = function(x) getLearnerModel(x),
                         measures = list(rmse, rsq), show.info = FALSE)

model_rf

Mostly, the model predicts coherent results that make sense. However, when I run exactly the same code (no changes whatsoever) on the computers of two of my colleagues, the model predicts those weird results:

Resample Result
Task: data.model
Learner: regr.randomForestSRC
Aggr perf: rmse.test.rmse=361.1464455,rsq.test.mean=-588.1729057
Runtime: 4.0032

What could be the reason for this weird behavior on only two computers but not the other ones?

Have you checked that you have the same versions of R, mlr, and libraries? — Lars Kotthoff, May 14 '18 at 15:11
Indeed, it is a version conflict. Not working with: randomForestSRC_2.6.0, but working with: randomForestSRC_2.5.1 — Jan Janiszewski, May 15 '18 at 09:20

score 0 · Accepted Answer · answered May 15 '18 at 09:21

0

This was a version conflict. After installing an older version of randomForestSRC, it all worked on my computer as well.

It definitely did not work with randomForestSRC version: 2.6.0

It definitely works now with randomForestSRC version: 2.5.1

answered May 15 '18 at 09:21

Jan Janiszewski

432
3
14

MLR and randomForestSRC: Incoherence between computers

1 Answers1