v1.5.4

Major updates:

1. Update Clip operator
2. Fix on optimization, unit test added.

v1.5.3

Major update:

1. Perform topological sorting on node list when we make onnx model.
2. Several changes on optimizer.

v1.5.0

onnxconverter-common version 1.5.0 is now available! This package was created alongside the onnxmltools package to enable common code to be shared among various ONNX converter offerings.

onnxmltools, sklearn-onnx, and keras-onnx depend on the onnxconverter-common package for shared code and utilities, including:

1. Converting common ONNX operations (onnx_ops.py)
2. ONNX graph topology and data types (topology.py, container.py, data_types.py)
3. Other shared tools, including the float32->float16 converter (float16.py), ONNX graph optimizer (optimizer.py) and common shape calculators (shape_calculator.py)

How do I use the latest onnxconverter-common package?

pip install onnxconverter-common
python -c "import onnxconverter_common"

This package is automatically installed as part of the dependencies for onnxmltools, keras2onnx, and skl2onnx.

Highlights since the last release

• v1.5.0 for the onnxconverter-common package (onnx/onnxmltools#314)
• [Opset 10] Updates for thresholded relu (onnx/onnxmltools#308)
• [Opset 10] Deprecate Upsample, create Resize op (onnx/onnxmltools#303)
• [Opset 10] Pooling operator updates: AveragePool, MaxPool (onnx/onnxmltools#296)
• Add support for Py3.7, onnx 1.5, onnxruntime 0.4 (onnx/onnxmltools#293)
• Added apply_slice function to enable multiple versions of Slice (onnx/onnxmltools#291)