Catboost support

[rspadim]

No description provided.

[hcho3]

@rspadim Are you familiar with how Catboot saves models?

[rspadim]

there're some types, and very interesting since some are wellknown formats (standards)
i will point it here, and try to upload some samples
https://github.com/catboost/catboost/blob/master/catboost/libs/model/model.cpp#L174

[rspadim]

well... now i know that it can export to python and c++ too

https://github.com/catboost/tutorials/blob/7897e65888889f559715bf0da1222a9b6dbc9d6b/apply_model/model_export_as_cpp_code_tutorial.md

https://github.com/catboost/tutorials/blob/7897e65888889f559715bf0da1222a9b6dbc9d6b/apply_model/model_export_as_python_code_tutorial.md

sorry my fault

[hcho3]

@rspadim It may be still beneficial to add Catboost support in Treelite, if you'd like to generate pure C code without external dependency. That said, closing for now.

[rspadim]

an exampel to create models:

from sklearn.datasets import load_boston
X, y = load_boston(return_X_y=True)
print('dimensions of X = {}'.format(X.shape))
print('dimensions of y = {}'.format(y.shape))

from catboost import CatBoost
model = CatBoost()
model.fit(X, y)
for i in ['cpp', 'python', 'json', 'cbm', 'onnx', 'coreml']:
    print(i)
    model.save_model(fname="teste.catboost." + i, format=i)
print("hello")

[rspadim]

hi @hcho3 , yes i agree with you, pure C is the best think treelite have, it's very intuitive to port to any language

[hcho3]

@rspadim It is the job of Treelite contributors to support Catboost, not Catboost contributors. I think you should close catboost/catboost#721.

Also re-opening this issue, as a reminder that we should support Catboost

[rspadim]

Nice, reference to create treelite support

https://github.com/catboost/catboost/blob/6ec9193d0cbb9a8712bd8fe434b823f30c4ee875/catboost/libs/model/model.cpp#L164

https://github.com/catboost/catboost/tree/15712cfa704413d51618455326c30f5764956be5/catboost/libs/model/model_export

cpp might be easy to port

[hcho3]

Closing in favor of #377