Converting XGBoost to ONNX

[dsmanl]

Dear all,

first of all thanks a lot for all of your efforts.

I am currently trying to convert a XGBoost classifier to ONNX format using a toy example. While doing that I am getting the following error AttributeError: 'numpy.uint8' object has no attribute 'encode'.

I am running on macOS Catalina, Version 10.15.6 and using the following Python (3.7.7) packages:

• xgboost==1.1.0
• onnx==1.6.0
• onnxconverter-common==1.7.0
• onnxmltools==1.7.0

The minimum working example is:

import tensorflow as tf
from pathlib import Path

from sklearn.preprocessing import StandardScaler
import xgboost as xgb
import onnxmltools

def xgboost_example():
	"""
	Train a simple xgboost model and store associated artefacts.
	"""
	path = Path('/PATH/TO/LOCAL/FOLDER')

	(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()

	X_train = X_train.reshape((X_train.shape[0], -1))
	#X_test = X_test.reshape((X_test.shape[0], -1))

	scaler = StandardScaler()

	X_train = scaler.fit_transform(X_train)
	#X_test = scaler.transform(X_test)

	sh = X_train[0].reshape(1, -1).shape

	clf = xgb.XGBClassifier(objective="multi:softprob", n_jobs=-1)
	clf.fit(X_train, y_train)

	onnx_model = onnxmltools.convert_xgboost(clf, initial_types=[
		('float_input', onnxmltools.convert.common.data_types.FloatTensorType(sh))])
	onnxmltools.utils.save_model(onnx_model, path.joinpath('model_onnxmltools.onnx'))


if __name__ == "__main__":
	xgboost_example()

The traceback is:

Traceback (most recent call last):
  File "test_xgb.py", line 40, in <module>
    xgboost_example()
  File "test_xgb.py", line 35, in xgboost_example
    ('float_input', onnxmltools.convert.common.data_types.FloatTensorType(sh))])
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxmltools/convert/main.py", line 85, in convert_xgboost
    return convert(*args, **kwargs)
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxmltools/convert/xgboost/convert.py", line 47, in convert
    onnx_model = convert_topology(topology, name, doc_string, target_opset, targeted_onnx)
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxconverter_common/topology.py", line 776, in convert_topology
    get_converter(operator.type)(scope, operator, container)
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxmltools/convert/xgboost/operator_converters/XGBoost.py", line 287, in convert_xgboost
    cls.convert(scope, operator, container)
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxmltools/convert/xgboost/operator_converters/XGBoost.py", line 248, in convert
    classes = np.array([s.encode('utf-8') for s in classes])
  File "/usr/local/anaconda3/envs/machine-learning/lib/python3.7/site-packages/onnxmltools/convert/xgboost/operator_converters/XGBoost.py", line 248, in <listcomp>
    classes = np.array([s.encode('utf-8') for s in classes])
AttributeError: 'numpy.uint8' object has no attribute 'encode'

Do you have any ideas on whether I am doing something wrong or whether there is another issue?

Thanks a lot!