OVH Guides

Exporting Scikit-learn models

Learn how to export a Scikit-learn model through ONNX format

Last updated 7th February, 2020.


Scikit-learn is a popular machine learning library and ONNX is a serialization format that is supported by OVHcloud Serving Engine. This tutorial will cover how to export a Scikit-learn trained model into an ONNX file.


Convert a simple model into ONNX

Serving Engine supports scikit-learn models through the ONNX serialization format.

Train Simple scikit-learn model

Let\'s take a simple example of a scikit-learn model to illustrate:

# Train a model.
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier

iris = load_iris()
X, y =,
X_train, X_test, y_train, y_test = train_test_split(X, y)
classifier = RandomForestClassifier(), y_train)

Install sklearn-onnx module

Within your scikit-learn project, just install sklearn-onnx using PyPi:

pip install skl2onnx

Define the inputs of your serialized model

For each numpy array (also called tensor in ONNX) fed as an input to the model, choose a name and declare its data-type and its shape.


# import needed data type
from skl2onnx.common.data_types import FloatTensorType
# input tensors of your model: list of ('<wanted name of tensor>', DataType('<shape>'))
initial_type = [
    ('float_input', FloatTensorType([None, 4]))

Launch the conversion and save it to a file

The trained model conversion is made with the convert_sklearn function.

# Import export function
from skl2onnx import convert_sklearn
# Export the model
onx = convert_sklearn(classifier, initial_types=initial_type)
# Save it into wanted file
with open("my_model.onnx", "wb") as f:

Your model is now serialized on you local file system in the my_model.onnx file.

Going further

These guides might also interest you...