How to convert a trained LGBM model to DAAL model by CPP interfaces?

oneapi-src / oneDAL

oneAPI Data Analytics Library (oneDAL)

https://software.intel.com/en-us/oneapi/onedal

Apache License 2.0

606 stars 213 forks source link

How to convert a trained LGBM model to DAAL model by CPP interfaces? #2836

Open AllenSun1024 opened 3 months ago

AllenSun1024 commented 3 months ago

Describe the bug

In Python, we can simply use daal4py.get_gbt_model_from_lightgbm() to convert a trained LGBM model to oneDAL, and use daal4py.gbt_regression_prediction() to predict.
May I find the corresponding CPP interfaces to convert models and predict?

razdoburdin commented 3 months ago

Hi, the conversion has only python implementation for now.

AllenSun1024 commented 3 months ago

Hi, @razdoburdin

Thanks for your kind reply.

Further, have your team started to implement the CPP converter interfaces?

Best, Chao

razdoburdin commented 3 months ago

Hi, @razdoburdin

Thanks for your kind reply.

Further, have your team started to implement the CPP converter interfaces?

Best, Chao

Could you please describe your use case? We don't have such plans for now.

AllenSun1024 commented 2 months ago

Hi, @razdoburdin

My use case is described below:

I have a LightGBM model trained through LightGBM's Python API
I use LightGBM's C++ API to load the pre-trained model and infer
To make the speed of inference faster, I need some C++ APIs which are similar to Python daal4py to convert the LightGBM model to DAL model.

Best, Chao

syakov-intel commented 2 months ago

Hi @AllenSun1024 , On the last stage, do you have particular requirement/expectation? Is your case latency sensitive? Could you please elaborate more?

AllenSun1024 commented 2 months ago

Hi, @syakov-intel

My case is latency sensitive. I wanna use oneDAL to accelerate on Intel Xeon CPU. I think inferring on Intel Xeon CPU with oneDAL's C++ interfaces will make a great progress in speed just like Python daal4py, right?