[Roadmap] 1.3.0 Roadmap

[trivialfis]

@dmlc/xgboost-committer Please add your items here by editing this post. Let's ensure that

• Each item has to be associated with a ticket
• Major design/refactoring are associated with a RFC before committing the code
• Blocking issue must be marked as blocking
• Breaking change must be marked as breaking

For other contributors who have no permission to edit the post, please comment here about what you think should be in 1.3.0.

• Merge rabit into XGBoost ([RFC] Merging RABIT into XGBoost. #5995, Merge rabit #6023)
• GPU Shap value compuation. (GPUTreeShap #6038, Update GPUTreeShap #6064, Update GPUTreeshap #6163)
• Let Spark package fail gracefully (XGBoost-Spark killing SparkContext on Task Failure #4826, [jvm-packages] cancel job instead of killing SparkContext #6019)
• Support for more different meta on dask. ([dask] Refactor meta data handling. #6130, [dask] Support more meta data. #6132)
• Support MAPE. (Add MAPE metric #6119)
• GPU predict leaf function (GPU Support for pred_leaf #5847, Implement GPU predict leaf. #6187)
• New callback functions ([POC] Implement callback for dask. #5612, Rework Python callback functions. #6199)
• Early stopping on dask. (Rework Python callback functions. #6199)
• Initial support for categorical data. (Initial support for one hot split. #5949, Support categorical data in GPU sketching. #6137, Support categorical data in ellpack. #6140, Add categorical data support for gpu_hist updater. #6164, Add categorical data support to GPU predictor. #6165, Enable categorical data support on Python DMatrix. #6166, High level tests for categorical data. #6179, Add JSON schema for categorical splits. #6194, More categorical tests and disable shap sparse test. #6219)
• Feature weights for column sampling (exact, hist, gpu_hist) (Feature weights #5962)
• Slicing model. (Removing trees from a model #5531, [feature proposal] Load best model after early stopping #4052, Support slicing tree model #6302)
• Default to JSON for memory snapshot. (Make JSON the default full serialization format. #6027)
• Support dask on reverse proxy environment. (Host IP resolution problem w/ dask on kubernetes or dask-gateway #5765, [dask] Accept user defined tracker address. #5408, [dask] Move RABIT tracker to first worker #6142, [dask] Fix running on k8s. #6343)
• Support using system installation of libxgboost.so when distributing Python package. (Support shared library in system path. #6362)

[trivialfis]

Moving:

• Support dask on reverse proxy environment. (Host IP resolution problem w/ dask on kubernetes or dask-gateway #5765, [dask] Accept user defined tracker address. #5408, [dask] Move RABIT tracker to first worker #6142)

off the list again. Getting it work on GKE is a bit tricky. Related:

• dask kubernetes: https://kubernetes.dask.org/en/latest/
• kubernetes
• docker image deployment
• GKE
• RABIT internal

[trivialfis]

The item is moved back in now that I can run tests myself.

[trivialfis]

RC1 is targeted at Nov 20.