Unsupported objective 'quantile' for LGBMRegressor

[brunovilar]

Issue description

When trying to run onnxmltools.convert_lightgbm on a lightgbm.sklearn.LGBMRegressor model with objective='poisson', the conversion fails with a RuntimeError.

Open PR

This is an update. O just saw this pull request. By directly applying the change it works properly. I would really recommend approving and merge the PR. :)

Similar closed issue

This problem seems to be similar to this one, reported by @janjagusch. I could not reproduce the workaround for the original (poisson) problem.

Reproducible example

import numpy as np
from lightgbm.sklearn import LGBMRegressor
from onnxmltools import convert_lightgbm
from skl2onnx.common.data_types import FloatTensorType

examples = 1000
features = 10

X = np.random.randn(examples, features)
y = np.round(np.random.randn(examples), 0)

model = LGBMRegressor(objective="quantile", metric="quantile", alpha=0.05)
model.fit(X, y)

initial_types = [("float_input", FloatTensorType([None, features]))]
onnx_model = convert_lightgbm(model, initial_types=initial_types)

Results in the following error:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_33543/973951346.py in <module>
     14 
     15 initial_types = [("float_input", FloatTensorType([None, features]))]
---> 16 onnx_model = convert_lightgbm(model, initial_types=initial_types)

/media/bruno/dados/dev/virtual_envs/generic_env/lib/python3.9/site-packages/onnxmltools/convert/main.py in convert_lightgbm(model, name, initial_types, doc_string, target_opset, targeted_onnx, custom_conversion_functions, custom_shape_calculators, without_onnx_ml, zipmap)
    134 
    135     from .lightgbm.convert import convert
--> 136     return convert(model, name, initial_types, doc_string, target_opset, targeted_onnx,
    137                    custom_conversion_functions, custom_shape_calculators, without_onnx_ml,
    138                    zipmap=zipmap)

/media/bruno/dados/dev/virtual_envs/generic_env/lib/python3.9/site-packages/onnxmltools/convert/lightgbm/convert.py in convert(model, name, initial_types, doc_string, target_opset, targeted_onnx, custom_conversion_functions, custom_shape_calculators, without_onnx_ml, zipmap)
     54                               custom_shape_calculators, zipmap=zipmap)
     55     topology.compile()
---> 56     onnx_ml_model = convert_topology(topology, name, doc_string, target_opset, targeted_onnx)
     57 
     58     if without_onnx_ml:

/media/bruno/dados/dev/virtual_envs/generic_env/lib/python3.9/site-packages/onnxconverter_common/topology.py in convert_topology(topology, model_name, doc_string, target_opset, targeted_onnx, channel_first_inputs)
    774         else:
    775             # Convert the selected operator into some ONNX objects and save them into the container
--> 776             get_converter(operator.type)(scope, operator, container)
    777 
    778     # When calling ModelComponentContainer's add_initializer(...), nothing is added into the input list.

/media/bruno/dados/dev/virtual_envs/generic_env/lib/python3.9/site-packages/onnxmltools/convert/lightgbm/operator_converters/LightGbm.py in convert_lightgbm(scope, operator, container)
    398         post_transform = "Exp"
    399     else:
--> 400         raise RuntimeError(
    401             "LightGBM objective should be cleaned already not '{}'.".format(
    402                 gbm_text['objective']))

RuntimeError: LightGBM objective should be cleaned already not 'quantile'.

Versions

numpy==1.21.1
lightgbm==3.2.1
onnxmltools==1.9.1
skl2onnx==1.10.0

[brunovilar]

As the PR was merged, this issue is closed.