lightgbm with categorical feature error

package version: onnx: 1.8.1 onnxruntime: 1.4.0 skl2onnx: 1.8.0 onnxmltools: 1.7.0

here is my code

` import pandas as pd from sklearn.model_selection import train_test_split import onnx import onnxmltools

import lightgbm as lgb from lightgbm import LGBMClassifier from skl2onnx.common.data_types import FloatTensorType from onnxmltools.convert.lightgbm.operator_converters.LightGbm import convert_lightgbm # noqa from skl2onnx.common.shape_calculator import calculate_linear_classifier_output_shapes # noqa from skl2onnx import convert_sklearn, update_registered_converter, version import onnxruntime as rt

print("onnx: ", onnx.version) print("onnxruntime: ", rt.version) print("skl2onnx: ", version) print("onnxmltools: ",onnxmltools.version)

CATE_MODE = True sample_name=‘sample2.csv’ cate_col_str = ‘mid,204,2591,2603’

parameters = { ‘n_jobs’: 1, ‘objective’: ‘binary’, ‘metric’: ‘auc’, ‘is_unbalance’: ‘true’, ‘boosting’: ‘gbdt’, ‘num_leaves’: 50, ‘num_trees’: 100, ‘feature_fraction’: 0.9, ‘bagging_fraction’: 0.8, ‘learning_rate’: 0.1, ‘min_data_in_leaf’: 500, ‘max_depth’: 15, ‘verbose’: 0, }

flat = pd.read_csv(sample_name, sep=‘\t’) cols = flat.columns

features = [] for i in cols: if i != ‘label’: features.append(i)

for i in flat.columns: flat[i] = flat[i].astype(float)

flat[‘label’] = flat[‘label’].astype(int)

if CATE_MODE: for cate_col in cate_col_str.split(‘,’): flat = flat[(flat[cate_col] >= 0) & (flat[cate_col] % 1 == 0)] flat[cate_col] = flat[cate_col].astype(‘category’)

label = [‘label’] x = flat[features] y = flat[label]

x, x_test, y, y_test = train_test_split(x, y, test_size=0.2, random_state=42, stratify=y) x_train, x_valid, y_train, y_valid = train_test_split(x, y, test_size=0.2, random_state=42, stratify=y) train_data = lgb.Dataset(x_train, y_train, free_raw_data=False) test_data = lgb.Dataset(x_test, label=y_test, free_raw_data=False) valid_data = lgb.Dataset(x_valid, label=y_valid, free_raw_data=False)

clf = LGBMClassifier(**parameters) clf.fit(X=x_train, y=y_train, eval_set=[(x_valid, y_valid)], early_stopping_rounds=10) print(‘best score’, clf.best_score_)

update_registered_converter( LGBMClassifier, ‘LightGbmLGBMClassifier’, calculate_linear_classifier_output_shapes, convert_lightgbm, options={‘nocl’: [True, False], ‘zipmap’: [True, False] })

version = ‘testtest’ onnx_name = f"lgb_skl_{version}.onnx"

dim = x_train.shape[1] initial_type = [(‘float_input’, FloatTensorType([None, dim]))] onx = convert_sklearn(clf, initial_types=initial_type, target_opset=12, options={id(clf): {‘zipmap’: False}})

print(dim)

with open(onnx_name, “wb”) as f: f.write(onx.SerializeToString()) print(‘success’) `

When I set CATE_MODE=False, ONNX can be saved successfully. When I set CATE_MODE=True, and a small dataset sample.csv, ONNX can be saved successfully. When I set CATE_MODE=True, with a large dataset, sample2.csv. Here is the error:

` ValueError Traceback (most recent call last) /usr/local/lib/python3.6/dist-packages/skl2onnx/common/_container.py in add_node(self, op_type, inputs, outputs, op_domain, op_version, name, **attrs) 540 node = make_node(op_type, inputs, outputs, name=name, –> 541 _dtype=dtype, **attrs) 542 except ValueError as e:

/usr/local/lib/python3.6/dist-packages/skl2onnx/proto/onnx_helper_modified.py in make_node(op_type, inputs, outputs, name, doc_string, domain, _dtype, **kwargs) 67 make_attribute(key, value, dtype=_dtype, domain=domain) —> 68 for key, value in sorted(kwargs.items())) 69 return node

/usr/local/lib/python3.6/dist-packages/skl2onnx/proto/onnx_helper_modified.py in <genexpr>(.0) 67 make_attribute(key, value, dtype=_dtype, domain=domain) —> 68 for key, value in sorted(kwargs.items())) 69 return node

/usr/local/lib/python3.6/dist-packages/skl2onnx/proto/onnx_helper_modified.py in make_attribute(key, value, dtype, domain, doc_string) 183 key, type(value), dtype, –> 184 [type(_) for _, __ in zip(value, range(0, 5))])) 185 else:

ValueError: You passed in an iterable attribute but I cannot figure out its applicable type, key=‘nodes_values’, type=<class ‘list’>, dtype=None, types=[<class ‘float’>, <class ‘float’>, <class ‘float’>, <class ‘float’>, <class ‘float’>].

The above exception was the direct cause of the following exception:

ValueError Traceback (most recent call last) <ipython-input-29-e6c46f777ebd> in <module> 1 model_def = to_onnx(clf, x_train.values.astype(numpy.float32), ----> 2 options={id(clf): {‘zipmap’: False}})

/usr/local/lib/python3.6/dist-packages/skl2onnx/convert.py in to_onnx(model, X, name, initial_types, target_opset, options, white_op, black_op, final_types, dtype) 212 name=name, options=options, 213 white_op=white_op, black_op=black_op, –> 214 final_types=final_types, dtype=dtype) 215 216

/usr/local/lib/python3.6/dist-packages/skl2onnx/convert.py in convert_sklearn(model, name, initial_types, doc_string, target_opset, custom_conversion_functions, custom_shape_calculators, custom_parsers, options, intermediate, white_op, black_op, final_types, dtype) 160 onnx_model = convert_topology(topology, name, doc_string, target_opset, 161 options=options, –> 162 remove_identity=not intermediate) 163 164 return (onnx_model, topology) if intermediate else onnx_model

/usr/local/lib/python3.6/dist-packages/skl2onnx/common/_topology.py in convert_topology(topology, model_name, doc_string, target_opset, channel_first_inputs, options, remove_identity) 1085 type(getattr(operator, ‘raw_model’, None)))) 1086 container.validate_options(operator) -> 1087 conv(scope, operator, container) 1088 1089 # Create a graph from its main components

/usr/local/lib/python3.6/dist-packages/skl2onnx/common/_registration.py in call(self, *args) 27 if args[1].raw_operator is not None: 28 args[2]._get_allowed_options(args[1].raw_operator) —> 29 return self._fct(*args) 30 31 def get_allowed_options(self):

/usr/local/lib/python3.6/dist-packages/onnxmltools/convert/lightgbm/operator_converters/LightGbm.py in convert_lightgbm(scope, operator, container) 300 ‘TreeEnsembleClassifier’, operator.input_full_names, 301 [label_tensor_name, probability_tensor_name], –> 302 op_domain=‘ai.onnx.ml’, **attrs) 303 304 prob_tensor = probability_tensor_name

/usr/local/lib/python3.6/dist-packages/skl2onnx/common/_container.py in add_node(self, op_type, inputs, outputs, op_domain, op_version, name, **attrs) 542 except ValueError as e: 543 raise ValueError(“Unable to create node ‘{}’ with name=‘{}’.” –> 544 “”.format(op_type, name)) from e 545 node.domain = op_domain 546

ValueError: Unable to create node ‘TreeEnsembleClassifier’ with name=‘LightGbmLGBMClassifier’. `

Also, When I change ‘num_trees’ from 100 to 10(CATE_MODE=True, with a large dataset), the error disappear.

[xadupre]

Your example don’t fail anymore with the latest changes in skl2onnx and onnxmltools (version 1.9.0, to be release soon, available but available on github).

[xadupre]

I’ll close the issue. Feel free to reopen if needed.