Rborist segfaults on Mac when run rapidly in succession

[stevenbagley]

On Mac OSX 10.12.6, R 3.4.3, Rborist 0.1-8, use the following code chunk:

library(Rborist)
data(iris)

run <- function(n){
    for(i in 1:n){
        cat(i, " ")
        validation_index <- sample(1:nrow(iris), 30)
        validation_data <- iris[-validation_index,]
        training_data <- iris[validation_index,]
        rb <- Rborist(training_data[, -5], training_data[, 5])
    } }

then: run(100)

will run 10 or 20 times, then seg fault:

> run(100)
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  
 *** caught segfault ***
address 0x7fc7561143c0, cause 'memory not mapped'

Traceback:
 1: doTryCatch(return(expr), name, parentenv, handler)
 2: tryCatchOne(expr, names, parentenv, handlers[[1L]])
 3: tryCatchList(expr, classes, parentenv, handlers)
 4: tryCatch(.Call("RcppTrainCtg", predBlock, preFormat$rowRank,     y, nTree, nSamp, rowWeight, withRepl, treeBlock, minNode,     minInfo, nLevel, maxLeaf, predFixed, splitQuant, probVec,     autoCompress, thinLeaves, FALSE, classWeight), error = function(e) {    stop(e)})
 5: Rborist.default(training_data[, -5], training_data[, 5], classWeight = 1:3)
 6: Rborist(training_data[, -5], training_data[, 5], classWeight = 1:3)
 7: run(100)

[suiji]

Thank you for providing a test case. Will try to reproduce.

[suiji]

Reproduces with 0.1-9 under Fedora 26.

Investigating.

[suiji]

This is an error in the splitting method for compressed predictor values. Until a fix is released, it should be possible to work around the error by turning off autocompression. This can be done by specifying "autoCompress=1.0".

[suiji]

Appears to be repaired. One of the sparse splitting methods employed a bad initialization. These methods are somewhat fragile, and should be reimplemented more cleanly.

"run(10000)" now completing without incident.

Please feel free to reopen this issue, as needed. Again, thank you for isolating the problem.