In general, the current project is still under early development. It is expected that a full usable demo, along with a docker environment to be presented before 2020-11-14.
- Currently, the docker image cannot support rapids-ai yet. This is because, for some reason, inside the docker container, conda activate cannot be performed. Someone needs to fix this problem.
- None of the encoders have been fully tested. Please use the data in examples to perform the test.
- In tabular/model_fitter.py, it is helpful for this Opt (such as LGBOpt) to add a
__repr__methods - In tabular/model_fitter.py, a seed option should be added for each model for reproduction.
- In tabular/model_fitter.py, such as in the LGBFitter.train method, some options given be hyperopt might work well together. Therefore some guards should be placed.
- In tabular/encoders, some encoders should only be applied to certain types of variables. For example,
category variables should only be applied to variables that starts with discrete. A warning (using
warning.warn())) should be added. - In tabular/model_fitter.py, cuml fitter's train_k_fold methods should return, at the fourth position, the trained models.
- Please complete the docstring for all the encoder's configs.
- All the model only supports binary classification now. It would be nice to add options for other types of targets.
- Add support to merge lamb into RAdamW.
Among almost everything.
- All the optimizers utilities are not tested.