[Draft] Oct~Nov iteration Plan

[scarlett2018]

We'd like to experience the practices to use issues tracking recent planned items. This issue is to cover recent updates and changes in the current & coming months, all are welcomed to comments/add in the items you'd like to merge into the iteration. Items by area:

PRs to be finished in this iteration

• P0 - Target and Count encodings for categorical features #3234 @shiyu1994
• P0 - [CUDA] New CUDA version Part 1 #4630 @shiyu1994
• P1 - Minimal Varianсe Sampling booster #4266

Feature requests decomposed from the above PRs

• P0 - Target and Count Encoding for Categorical Features with Ensemble #4779
• P1 - Support Target and Count Encoding for Sequence Input #4780 @hzy46 break down items
• P1 - [CUDA] Monotonic Constraints on New CUDA Version #4781
• P1 - [CUDA] Feature Interaction Constraints on New CUDA Version #4785

Bug fixes and small improvement

• ignore_column does not work when loading a dataset from a binary file ignore_column does not work when loading a dataset from a binary file #4657
• Memory leaks / is better memory management possible? Memory leaks / is better memory management possible? #4239
• Using feature_fraction and forced splits together can cause training failures Using feature_fraction and forced splits together can cause training failures #4601
• Passing default num_threads to booster.predict not working Passing default num_threads to booster.predict not working #4607
• [python] suppress the warning about categorical feature override [python] suppress the warning about categorical feature override #3379
• trees with one leaf can have strong bias in predictions, far outside observed range trees with one leaf can have strong bias in predictions, far outside observed range #4708
• Allow user to ignore min_gain_to_split when using forced splits by adding a hyper parameter (ForceSplit might be ignored if no gain #4726)
• Avoiding Exception "Check failed: (best_split_info.right_count) > (0) at ..." with a regression task (Avoiding Exception "Check failed: (best_split_info.right_count) > (0) at ..." with a regression task #3679)
• simple repro for lightgbm.basic.LightGBMError: Check failed: (best_split_info.left_count) > (simple repro for lightgbm.basic.LightGBMError: Check failed: (best_split_info.left_count) > (0) #4739 )

End to End Tests

• consider to leverage the related project: https://github.com/Microsoft/LightGBM-benchmark

[shiyu1994]

An on-going work for benchmark of LightGBM
https://github.com/Microsoft/LightGBM-benchmark

[hzy46]

(Candidate) Bugs to be fixed:

• ignore_column does not work when loading a dataset from a binary file #4657
• Memory leaks / is better memory management possible? #4239
• Using feature_fraction and forced splits together can cause training failures #4601
• Passing default num_threads to booster.predict not working #4607

[tongwu-sh]

List of dependencies that should be updated from time to time

[StrikerRUS]

@shiyu1994 @guolinke I see there are new guys with member and collaborator statuses in the LightGBM repository. This is awesome! 🚀 I think it'll good to add them into FAQ and CODEOWNERS sections (refer to #4563 or #4060 as an example).

[shiyu1994]

@StrikerRUS Yes, that would be great. I'll add that later today.

[shiyu1994]

A new item #4708 is added to the list.

[shiyu1994]

Some work items for the two large PRs #3234 and #4630 are separated as feature requests. See #4779 #4780 #4781 and #4785. Contributions are welcome if you are interested @hzy46 @tongwu-msft!

[StrikerRUS]

Today is March, so closing this plan as stale. Feel free to transfer some content from this issue into a new one if needed.

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed.
To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues
including a reference to this.