IVaps.helpers

Helper functions

Functions

check_conversion(model_path, …[, …])	Check successful conversion of ONNX model
convert_data_to_pb(pickle_path[, …])	Convert pickle test data file to ONNX .pb files.
convert_to_onnx(model, framework[, …])	Convenience function to quickly convert and save ONNX model with expected input/output settings
cumMean1D(X, S)	Return mean of every S rows
cumMean2D(X, S)	Return mean of every S rows for every delta
run_onnx_session(inputs, sess, input_names)	Convenience function to execute ONNX inference with an optional post-inference function
standardize(X)	Standardize 2D array of variables

IVaps.helpers.check_conversion(model_path: str, onnx_model_path: str, framework: str, test_input_path: Optional[str] = None, tf_input_names: Optional[Sequence] = None, tf_output_names: Optional[Sequence] = None, log_path: Optional[str] = None)

Check successful conversion of ONNX model

Parameters

• model_path (str) – Path to original saved model

• onnx_model_path (str) – Path to converted ONNX model

• framework (str) – Reference string for one of the implemented frameworks

• test_input_path (str, default: None) – Path to folder with saved .pb test inputs

• tf_input_names (Sequence, default: None) – Names of inputs for Tensorflow model, if applicable

• tf_output_names (Sequence, default: None) – Names of outputs for Tensorflow model, if applicable

• log_path (str, default: None) – Path to save test results

Returns

True if model passses all checks

Return type

bool

IVaps.helpers.convert_data_to_pb(pickle_path: str, output_folder: str = 'test_data_set_0', is_input=True)

Convert pickle test data file to ONNX .pb files.

Parameters

• pickle_path (str) – The path to your pickle file. The pickle file should contain a dictionary with the following format: { input_name_1: test_data_1, input_name_2: test_data_2, … }

• output_folder (str, default: “test_data_set_0”) – The folder to store .pb files. The folder should be empty and its name starts with test_data_*.

IVaps.helpers.convert_to_onnx(model, framework: str, dummy_input1=None, dummy_input2=None, output_path: Optional[str] = None, input_names: Tuple[str, str] = ('c_inputs', 'd_inputs'), output_names: Optional[Sequence] = None, tf_input_names: Optional[Sequence] = None, tf_output_names: Optional[Sequence] = None, target_opset: Optional[int] = None, **kwargs)

Convenience function to quickly convert and save ONNX model with expected input/output settings

Parameters

• model (object) – fitted model object (or path to saved model in Tensorflow case)

• framework (str) – Reference string for one of the implemented frameworks

• dummy_input1 (list-like, default: None) – Dummy input for first model input used for type inference and passed into downstream conversion functions

• dummy_input2 (list-like, default: None) – Dummy input for second model input (if applicable) used for type inference and passed into downstream conversion functions

• output_path (str, default: None) – path to save ONNX model

• input_names (Tuple[str, str], default: (“c_inputs”, “d_inputs”)) – input names to assign ONNX model

• output_names (list-like, default: None) – output names for later ONNX inference; if None defaults to naming the outputs sequentially “output_1”, “output_2”, etc…

• tf_input_names (list-like, default: None) – Input names for Tensorflow graph. Only required when converting from Tensorflow using a frozen graph or checkpoints.

• tf_output_names (list-like, default: None) – Output names for Tensorflow graph. Only required when converting from Tensorflow using a frozen graph or checkpoints.

• **kwargs (keyword arguments to be passed into mltools conversion function)

Returns

Converted ONNX model or boolean flag indicating successful conversion, depending on specific framework.

Return type

Object

IVaps.helpers.cumMean1D(X, S)

Return mean of every S rows

IVaps.helpers.cumMean2D(X, S)

Return mean of every S rows for every delta

IVaps.helpers.run_onnx_session(inputs: Sequence[numpy.ndarray], sess: onnxruntime.capi.onnxruntime_inference_collection.InferenceSession, input_names: Sequence[str], label_names: Optional[Sequence[str]] = None, fcn=None, **kwargs)

Convenience function to execute ONNX inference with an optional post-inference function

Parameters

• inputs (Sequence of array-likes) – ONNX inference inputs

• sess (onnxruntime InferenceSession)

• input_names (Sequence of strings) – Input names to assign to inputs

• label_names (Sequence of strings, default: all outputs) – Specific outputs to return from inference

• fcn (Object, default: None) – Vectorized function to pass inference outputs through

• **kwargs (additional arguments to pass into fcn)

Returns

Outputs of ONNX inference or post-inference function

Return type

np.ndarray

IVaps.helpers.standardize(X)

Standardize 2D array of variables