Code Refactoring

[guolinke]

Thanks to the active community, LightGBM receives many PRs for the new features.
Consequently, some codes become missing the readability and maintainability, for the coupling of many different functions.
For the convenience of future developments and friendliness for new users, I would like to have a code refactoring.

ping @chivee @StrikerRUS @jameslamb @Laurae2 @huanzhang12 @henry0312

you can add the files, which you think needs to refactor, to this issue.

• Cost efficient gradient boosting
• merge the logic of subset and subfeature
• remove the macro functions, maybe convert them to template functions
• refine the tree learning pipelines
• refine prediction codes
• remove the duplicate codes in **_bin

[StrikerRUS]

@guolinke Can we tick serial_tree_learner.cpp? Or do you want to do more refactoring?

[lcsdavid]

May I add that there is an error in
<LightGBM/include/LightGBM/dataset_loader.h>:30:26-48
CostructFromSampleData should be ConstructFromSampleData ?

Regards.

[guolinke]

@lcsdavid I think it is a typo. welcome to create PR to fix it.

[StrikerRUS]

I'd like to propose refactoring additional config structures introduced in the initial CUDA support (I believe they can be replaced by ordinary config object)

LightGBM/include/LightGBM/cuda/vector_cudahost.h

Lines 16 to 34 in 1782fcb

	enum LGBM_Device {
	lgbm_device_cpu,
	lgbm_device_gpu,
	lgbm_device_cuda
	};
	enum Use_Learner {
	use_cpu_learner,
	use_gpu_learner,
	use_cuda_learner
	};
	namespace LightGBM {
	class LGBM_config_ {
	public:
	static int current_device; // Default: lgbm_device_cpu
	static int current_learner; // Default: use_cpu_learner
	};

#3160 (comment)

[jameslamb]

It's been a few years since the last comment on this issue. I think we should close it, what do you think @guolinke ?

[guolinke]

Sure, let us close it for now

[jameslamb]

thank you very much!