prog2onnx 0.0.1

Convert Progressive Learning models to ONNX

Introduction

proglearn-onnx converts ProgLearn models to ONNX. Once in the ONNX format, you can use tools like ONNX Runtime for high performance scoring. All converters are tested with onnxruntime.

Supported Converters

This version implements ProgLearn estimators that adhere to the sklearn-onnx API on registering custom converters

Name	Package	Supported
ClassificationProgressiveLearner	progressive_learner	Yes
LifelongClassificationForest	forest	Yes
LifelongClassificationNetwork	network	No

Installation

You can install from PyPi:

pip install prog2onnx

Note: There is a known backtracking issue in pip's dependency resolver that may significantly affect the time required to fetch the correct versions of the dependencies. A quick and easy fix is to add --use-deprecated legacy-resolver at the end of pip install .

Getting started

# Train a model using 3 tasks.
import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from proglearn.forest import LifelongClassificationForest

iris = load_iris()
X, y = iris.data, iris.target
X = X.astype(np.float32)
X_train, X_test, y_train, y_test = train_test_split(X, y)

clr = LifelongClassificationForest(default_n_estimators=10)

for _ in range(3):
	clr.add_task(X_train, y_train)

# Convert into ONNX format.
from prog2onnx import Prog2ONNX

p2o = Prog2ONNX(clr)

# Convert for task_ID = 0
onx = p2o.to_onnx(0)

# Validate ONNX model (Optional) 
p2o.validate()

# Save ONNX model to file (Optional) 
p2o.save("forest_iris_0.onnx")

# Compute the prediction with onnxruntime.
import onnxruntime as rt

sess = rt.InferenceSession(onx.SerializeToString(), providers=["CPUExecutionProvider"])
input_name = sess.get_inputs()[0].name
pred_onx = sess.run(None, {input_name: X_test.astype(np.float32)})[0]

Testing

Several scenarios are assessed in the form of separate tests using the Python's built in unittest testing framework.

python -m unittest -v tests/test_proglearn.py

Contribute

We are open to contributions in the form of feedback, ideas, or code.

License

Apache License v2.0