Implementation:Microsoft Onnxruntime Numpy Output Extraction

Metadata

Field	Value
Implementation Name	Numpy_Output_Extraction
Repository	Microsoft_Onnxruntime
Source Repository	https://github.com/microsoft/onnxruntime
Type	Pattern Doc
Language	Python
Domain	ML_Inference, Model_Optimization
Last Updated	2026-02-10
Workflow	Python_Inference_Pipeline
Pair	6 of 6

Overview

Pattern documentation for extracting and interpreting ONNX Runtime inference outputs using standard numpy array indexing on session.run() return values.

API Signature

results = session.run([output_name], {input_name: x})
predictions = results[0]  # Standard list/array indexing

Import

import numpy

Code Reference

Reference	Location
Usage example	docs/python/examples/plot_load_and_predict.py:L55

I/O Contract

Inputs

Parameter	Type	Required	Description
Results from `session.run()`	`list[numpy.ndarray]`	Yes	The list of numpy arrays returned by the inference session's run method.

Outputs

Output	Type	Description
predictions	`numpy.ndarray`	Individual output tensor extracted by list index.
predicted_class	`numpy.ndarray`	(For classification) Class labels derived via `numpy.argmax()`.
probabilities	`numpy.ndarray`	(For probabilistic models) Probability distributions over classes.

Usage Example

Basic Output Extraction

results = sess.run([output_name], {input_name: x})
predictions = results[0]  # First output tensor

Classification Post-Processing

results = sess.run([output_name], {input_name: x})
predictions = results[0]
# Extract predicted class label
predicted_class = numpy.argmax(predictions, axis=-1)

Multiple Output Extraction

# Request multiple outputs
label_name = sess.get_outputs()[0].name
prob_name = sess.get_outputs()[1].name
results = sess.run([label_name, prob_name], {input_name: x})
labels = results[0]       # Predicted labels
probabilities = results[1] # Probability distributions

From the source example at docs/python/examples/plot_load_and_predict.py:L54-55:

res = sess.run([output_name], {input_name: x})
print(res)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment