Alex Bertram

Renjin was initially based on R 2.14.2 but is now closer to GNU R 3.2.2. We've been trying to keep the core packages in sync with those from GNU R...

Looks like returning `NULL` is insufficient: http://packages.renjin.org/package/org.renjin.cran/gbm/2.1.1/build/234#test-print.gbm-examples Think it might be helpful to extract a test case from this package and add to [packages/parallel/src/test/R](https://github.com/bedatadriven/renjin/tree/master/packages/parallel/src/test/R).

Getting rid of `NumberFormat.format()` helps a bit, cutting down prediction runtime by about 50%. But the match() routine still could be _greatly_ improved, it's really a minimal, naive implementation that...

@hannesmuehleisen `match()` is basically a very simple join function. What kind of decision tree does MonetDB use to decide between sorting, hashing, and a nothing? Have something in mind like,...

I just took a second to split the benchmark above into training and prediction phases, and the `match` routine basically dominates the prediction phase: ``` Function Count Time Own Time...

Actually, digging in a little, most of the time is actually spent on this call: [match(row.names(x), rn)](https://github.com/cran/randomForest/blob/4.6-12/R/predict.randomForest.R#L67) Renjin actually has the advantage of knowing that `rn` is a sequence, so...

After the latest commit, for this benchmark, prediction runtime dropped from 24.5 seconds to 4.7 seconds. The `match()` call consumes basically no (121 ms) runtime now. ``` > Completed in...

@thorstenwagner would be interesting to add your trajectory prediction as a benchmark as well. It looks like the model is serialized already, so we could include that directly. Could you...

Renjin's package management works a bit differently (we hope better) than GNU R's, so I'm not sure the ctv package will be usable with Renjin, especially if it relies on...

Do you get some kind of error or stack trace?