Automate testing for GPT with mock implementation #21
Conversation
There was a problem hiding this comment.
@ByteYJ , I think the OpenAI mocks are off to a good start, but we should actually be testing our own code somehow with this. For example, we should add a test for key functions in src/msfocr/llm/ocr_functions.py, like get_results or parse_table_data. Without checking those functions, it's kind of just an example on how the OpenAI API works, which isn't really the purpose of a unit test. Remember that our goal is to test the code that we're writing and make sure they work correctly in normal situations (OpenAI returns a correct json response) and edge cases (OpenAI returns an error or malformed json). The thing you want to mock is the request to the OpenAI client and force it to return the response you need for the test.
Could you try to add a test for one of those functions that uses your mocked classes in one of those situations?
Some other things you might want to watch out for:
- You're installing the
pytest_mockdependency, but it's not used in the tests right now. Instead you're using the built-inunittestlibrary, which is an alternative to thepytestlibrary we're using for our tests. I think this is okay and they should work together, but I haven't tried it before, although there is a reference on https://docs.pytest.org/en/7.1.x/how-to/unittest.html. - While having logging is a good idea later on, we don't need logging in the unit test files, that can be removed.
|
@ginic thanks for the suggestions! I added more tests to ocr_functions. now it includes six test cases. |
| with Image.open(image_path) as image: | ||
| image.load() | ||
|
|
||
| orientation = None | ||
| try: | ||
| for orientation in ExifTags.TAGS.keys(): | ||
| if ExifTags.TAGS[orientation] == 'Orientation': | ||
| break | ||
| exif = dict(image.getexif().items()) | ||
| if exif.get(orientation) == 3: | ||
| image = image.rotate(180, expand=True) | ||
| elif exif.get(orientation) == 6: | ||
| image = image.rotate(270, expand=True) | ||
| elif exif.get(orientation) == 8: | ||
| image = image.rotate(90, expand=True) | ||
| except (AttributeError, KeyError, IndexError): | ||
| pass | ||
| return image |
There was a problem hiding this comment.
Just a heads up, I was wrong about the correct way of handling the image earlier. I changed it again and I think this is correct now. You can read more at https://pillow.readthedocs.io/en/latest/reference/open_files.html#file-handling
There was a problem hiding this comment.
Thanks, @ByteYJ ! This look great! I'm really glad we have tests for these things now.
If you wanted to take things one step further, you could use your OpenAI mocker to test the extract_text_from_image function, but I think that's going above and beyond, so only do that if you feel like you have time.
This updates automating testing for ocr functionality and mock api call, including tests for:
Part 1: Testing llm_ocr_function
test_parse_table_data : ensures correct extraction and conversion of table data into DataFrames.
test_rescale_image : verifies the image resizing functionality for both largest and smallest dimensions.
test_encode_image : checks the base64 encoding and decoding of images.
test_correct_image_orientation : validates the correction of image orientation based on EXIF data.
Part 2: Testing OpenAI API call
test_query_api : verifies successful API responses using mocked responses.
test_query_api_authentication_error : tests handling of authentication errors using mocked error responses.
(These tests do not include extract_text_from_image and related functions, which requires a real API key and focuses more on OpenAI's effectiveness testing.)