search

h2oai / xgboost

Scalable, Portable and Distributed Gradient Boosting (GBDT, GBRT or GBM) Library, for Python, R, Java, Scala, C++ and more. Runs on single machine, Hadoop, Spark, Flink and DataFlow
Other
44 stars 26 forks source link

Patch Rabit issue causing slowdown of XGBoost on OS X #71

Closed michalkurka closed 5 years ago

Pscheidl commented 5 years ago

@michalkurka Approved, definitely an improvement. Sleep deprivation is better to be avoided. :+1: :rofl: Are we waiting for dmlc/xgboost to provide the ultimate fix ?

michalkurka commented 5 years ago

@Pscheidl yes, this should be fixed in XGBoost but this will solve this issue for us. The only penalty we incur is when a multi-node model training is finished. This fixes issues with scoring regardless of the number of nodes because scoring in H2O is always single-node (per-chunk).

Makes me think we are good now.