Skip to content

Latest commit

 

History

History
157 lines (128 loc) · 6.81 KB

test_examples.md

File metadata and controls

157 lines (128 loc) · 6.81 KB

How to test OpenVINO™ Python API?

Building and environment

Instructions can be found in "Building the OpenVINO™ Python API".

Install the specific requirements file for testing:

python -m pip install -r openvino/src/bindings/python/requirements_test.txt

Make sure that Python libraries are added to the user environment variables:

export PYTHONPATH=PYTHONPATH:<openvino_repo>/bin/intel64/Release/python

Run OpenVINO™ Python API tests

For simplicity, all of these commands require to navigate to the main Python API folder first:

cd .../openvino/src/bindings/python/

To run OpenVINO Python API 2.0 tests:

pytest tests/

To run OpenVINO Python API 1.0 tests, use this command:

pytest tests_compatibility/

By default, tests are run on the CPU plugin. If you want to run them on a different plugin, you need to specify this environment variable:

export TEST_DEVICE=GPU

The pytest test framework enables you to filter tests with the -k flag.

pytest tests/test_runtime/test_core.py -k "test_available_devices"

Alternatively, the full name and path to the test case could be passed.

pytest tests/test_runtime/test_core.py::test_available_devices

To print test names and increase verbosity, use -v flag.

pytest tests/test_runtime/test_core.py -v

Tip: look at pytest's documentation for more useful tricks: https://docs.pytest.org/en/latest/

To run full test suite one can utilize tox command:

tox

Check the codestyle of Python API

There are two packages used in the project to check the codestyle of python code: mypy and flake8. Besides, OpenVINO™ uses a custom configuration file to exclude some strict rules.

To check the codestyle of the Python API 2.0, run the following commands:

python -m flake8 ./src/openvino/ --config=setup.cfg
python -m mypy ./src/openvino --config-file ./setup.cfg

To check the codestyle of the nGraph Python API, run the following commands:

python -m flake8 ./src/compatibility/ngraph/ --config=setup.cfg
python -m mypy ./src/compatibility/ngraph --config-file ./setup.cfg

To check the codestyle of the InferenceEngine Python API, run the following commands:

cd src/compatibility/openvino
python -m flake8 ./ --config=setup.cfg
python -m mypy ./ --config-file ./setup.cfg

It's recommended to run the mentioned codestyle check whenever new tests are added. This check should be executed from the main Python API folder:

python -m flake8 ./tests/ --config=setup.cfg

Writing OpenVINO™ Python API tests

Before start

Follow and complete Examples of OpenVINO™ Python API code.

Adding new test-case in the correct place

Let's add a new test for OpenVINO™ Python API.

First, the test should confirm that the new pybind11-based class of MyTensor is behaving correctly. Navigate to tests folder and create a new file test_mytensor.py that describes tests within it. Final path should be along the lines of:

tests/test_runtime/test_mytensor.py

Don't forget to include license on the top of each new file!

Note that name of the file is connected to the class/module to be tested. This is exactly why tests are structured in folders that are describing what tests are supposed to be there. Always add tests to correct places, new folders and files should be created only when necessary. Quick overview of the structure:

tests/test_frontend           <-- frontend manager and extensions
tests/test_runtime            <-- runtime classes such as Core and Tensor
tests/test_graph              <-- operators and their implementation
tests/test_onnx               <-- ONNX Frontend tests and validation
tests/test_transformations    <-- optimization passes for OV Models 

Writing of the test itself

Let's add a test case for new class. Start with imports and simple test of the creation of a class:

import pytest
import numpy as np 
import openvino as ov

def test_mytensor_creation():
    tensor = ov.MyTensor([1, 2, 3])

    assert tensor is not None

Rebuilding step is not necessary here as long as there are no updates to codebase itself. Run the test with:

pytest tests/test_runtime/test_mytensor.py -v

In actual tests it is a good pratice to parametrize them, thus making tests compact and reducing number of handwritten test cases. Additionally, adding checks for shared functions to the basic tests is a common technique. Let's replace the test with:

@pytest.mark.parametrize(("source"), [
    ([1, 2, 3]),
    (ov.Tensor(np.array([4, 5 ,6]).astype(np.float32))),
])
def test_mytensor_creation(source):
    tensor = ov.MyTensor(source)

    assert tensor is not None
    assert tensor.get_size() == 3

Run the tests, output should be similar to:

tests/test_runtime/test_mytensor.py::test_mytensor_creation[source0] PASSED                                                                                                                                    [ 50%]
tests/test_runtime/test_mytensor.py::test_mytensor_creation[source1] PASSED                                                                                                                                    [100%]

Notice that the test name is shared between cases. In a real-life pull request, all of the functionalities should be tested to ensure the quality of the solution. Always focus on general usage and edge-case scenarios. On the other hand, remember that excessive testing is not advised as it may result in duplicate test cases and impact validation pipelines. A good "rule-of-thumb" list of practices while adding tests to the project is:

  • Don't test built-in capabilities of a given language.
  • Common functions can be tested together.
  • Create test cases with a few standard scenarios and cover all known edge-cases.
  • Hardcode desired results...
  • ... or create reference values during runtime. Always use a good, thrust-worthy library for that!
  • Re-use common parts of the code (like multiple lines that create helper object) and move them out to make tests easier to read.

Difference between tests and tests_compatibility directories

Someone could notice two similar folders tests and tests_compatibility. First one is the desired place for all upcoming features and tests. Compatibility layer is only supported in specific cases and any updates to it should be explicitly approved by OpenVINO™ reviewers. Please do not duplicate tests in both directories if not necessary.

See also