[ANNOUCEMENT] 0.81 release planned on November 1, 2018

[hcho3]

We will release version 0.81 on October 15, 2018.

Blocking issues / PRs

• colsample_bytree seemingly not random #3523: colsample_bytree seemingly not random
• MultiThreading not working properly with version 0.71.2 #3543: Multi-threading not working properly with version 0.71.2
• XGBoost - hist + learning_rate decay memory usage #3579: hist + learning_rate decay memory usage
• Lanuch function for unifying CPU and GPU code. [Reopen] #3643: Lanuch function for unifying CPU and GPU code
• Multi-GPU support in GPUPredictor. #3738: Multi-GPU support in GPUPredictor
• #3643 breaks binary wheels on CPU-only machine #3746: Binary wheels are broken on CPU-only containers
• [Blocking][jvm-packages] For training data with group, empty RDD partition threw exception (#3749) #3750: For training data with group, empty RDD partition threw exception
• [Blocking][jvm-packages] fix issue when spark job execution thread cannot return before we execute first() #3758: fix issue when spark job execution thread cannot return before we execute first()
• [Blocking][jvm-packages] fix the early stopping feature #3808: fix the early stopping feature in XGBoost4J
• [MGPU] benchmark/benchmark_tree.py failed. #3809: benchmark/benchmark_tree.py failed
• [Blocking] [jvm-packages] Upgrade Scala to 2.11.12 to address CVE-2017-15288 #3816: Upgrade Scala to 2.11.12 to address security vulnerability
• [Blocking] "tree_method=approx" does not invoke the correct updater #3840: tree_method=approx invokes the wrong tree updater class

[hcho3]

@RAMitchell @trivialfis @canonizer Do we want to merge #3643 and #3738 before the 0.81 release? The release determines what pip install xgboost will install.

[RAMitchell]

Yes I think we can merge #3643 now and #3738 should be easily doable before October 15.

[hcho3]

Adding #3746 as blocking issue.

[hcho3]

Pushing back the release date to November 1 due to blocking issues.

[CodingCat]

for #3543 and #3579 do we have active PR?

[hcho3]

Not yet. Still investigating.

[hcho3]

#3579 is addressed by #3803.

[trivialfis]

@hcho3 Can we add #3809 to blocking? I really worry about multi-gpu.

[hcho3]

@trivialfis Done

[trivialfis]

@hcho3 Thanks.

[hcho3]

I couldn't reproduce the bug reported by #3543 on my end using EC2. So we will ignore it for time being. The 0.81 release will happen as planned on November 1.

@CodingCat

[hcho3]

One thing I want to get done is the proper CI for CUDA 10.0: 1) a new build worker with CUDA toolkit 10.0 will run unit tests and 2) wheels built with CUDA 8.0 would be checked so that it installs on machines with CUDA 10.0.

[maeda]

Hi @hcho3 , do you know if this version will be release today?

[hcho3]

Unfortunately, the release is being delayed by #3840. I'm working on it.

[hcho3]

I will go ahead and draft the release note for the new version.

[hcho3]

I think #3856 is well tested and should be part of the 0.81 release. You will see a PR for new release by end of today (Pacific Time).