xgboost not running in parallel

[rladeira]

I recently reinstalled my operational system (ubuntu 14.04 LTS) and, strangely, xgboost is not able to run in parallel anymore. Even though I set the nthreads parameter to the number of cores I have, the parallel processing is not working at all.

I have tested both R and python packages without success. Is there some dependency that I must install to be able to run the algorithm in parallel?

[edi-bice]

Did you rebuild xgboost? I noticed when using build.sh script it switches to building single-threaded xgboost if it runs into any problems at all. I now avoid that script and do make instead. That way I can see and fix the problems.

[dirceusemighini]

Hi @edi-bice I'm having problem to execute xgboost on multiple workstations, does this parallel means multiple threads or multiple workstations?

[edi-bice]

@dirceusemighini in this thread parallel means multiple threads on the same workstation, i.e. does xgboost make full use of multicore workstations.

[dirceusemighini]

Thanks @edi-bice I can run xgboost in paralllel if I follow the instructions of https://github.com/dmlc/xgboost/issues/1106#issuecomment-245727917 although I can't make it to run with more than 1 nWorkers, as described here

[vkgpt11]

I am also facing the same issue. It is taking lots of time please suggest some solution.

[why-not]

I have compiled XGBoost with multi-threading enabled. But it still doesn't run in multiple threads (I set nthreads=-1 on an 8 core machine, no luck)

[hcho3]

@why-not You should set nthread=0 to use all available cores.