Onnx Helper¶

class systemds.onnx_systemds.onnx_helper.NodeTree(nodes: [<class 'onnx.onnx_ONNX_REL_1_7_ml_pb2.NodeProto'>])¶

A simple class for representing a tree structure of nodes

remove_end_node(node: systemds.onnx_systemds.onnx_helper.TreeNode)¶

Removes the given end-node from the tree. Removing a non-existing or non end-node raises an exception. :param node: The node that shall be removed

class systemds.onnx_systemds.onnx_helper.PreparedValue(value_info: onnx.onnx_ONNX_REL_1_7_ml_pb2.ValueInfoProto, initializer: onnx.onnx_ONNX_REL_1_7_ml_pb2.TensorProto = None)¶

Class for preparing onnx value structures for writing them to the dml script

systemds.onnx_systemds.onnx_helper.get_graph_inputs_with_initializers(graph: onnx.onnx_ONNX_REL_1_7_ml_pb2.GraphProto) → [(<class ‘onnx.onnx_ONNX_REL_1_7_ml_pb2.ValueInfoProto’>, <class ‘onnx.onnx_ONNX_REL_1_7_ml_pb2.TensorProto’>)]¶

Returns all initialized inputs of the graph with their corresponding initializer.

Parameters

graph – the onnx-graph

Returns

list of tuples of (input, initializer)

systemds.onnx_systemds.onnx_helper.get_graph_inputs_without_initializers(graph: onnx.onnx_ONNX_REL_1_7_ml_pb2.GraphProto) → [<class ‘onnx.onnx_ONNX_REL_1_7_ml_pb2.ValueInfoProto’>]¶

Returns all inputs of the graph that have no associated initializer values.

Parameters

graph – the onnx-graph

Returns

list of uninitialized inputs

systemds.onnx_systemds.onnx_helper.get_value_info(graph: onnx.onnx_ONNX_REL_1_7_ml_pb2.GraphProto, name: str) → onnx.onnx_ONNX_REL_1_7_ml_pb2.ValueInfoProto¶

Searches the graph for the given name and returns the associated ValueInfo, if the name is not found None is returned.

Parameters

• graph – the onnx-graph that shall be searched

• name – the name of the value

Returns

the value-info or None if it is not found

systemds.onnx_systemds.onnx_helper.load_model(onnx_file: str) → onnx.onnx_ONNX_REL_1_7_ml_pb2.ModelProto¶

Loads the onnx file, checks the model and converts it to a common version if necessary.

Parameters

onnx_file –

Returns

the loaded onnx-model