fix: Do not apply nutriscore prediction if it does not match the current nutriscore of the product #772
Conversation
| from PIL import Image | ||
| from robotoff.utils import get_image_from_url, get_logger, http_session | ||
|
|
||
| image = get_image_from_url('https://world.openfoodfacts.org/images/products/802/509/314/0251/4.jpg', error_raise=False) |
There was a problem hiding this comment.
good idea, but you should put it in the repository instead (in a tests/data directory for example), we wont to avoid tests depending on an external service.
There was a problem hiding this comment.
An image with a nutriscore would be more cool ;-)
Like https://world.openfoodfacts.org/images/products/330/274/003/0949/front_fr.110.400.jpg
(There are plenty, see eg https://world.openfoodfacts.org/label/nutriscore-grade-c)
| ocr_url = 'https://world.openfoodfacts.org/images/products/802/509/314/0251/4.json' | ||
| sever_domain = 'http://openfoodfacts.org/' | ||
|
|
||
| expected_predictions_all = get_predictions_from_image(barcode, image, source_image, ocr_url, sever_domain) |
There was a problem hiding this comment.
It's strange that you call it, to mock it !
In test you won't have a model.
Instead you should create a list of Prediction using PredictionFactory.
You are a bit lost maybe because you don't know what to put inside it, but what you are testing should tell you what to put in it. I can also provides you some data from production.
| ocr_url = 'https://world.openfoodfacts.org/images/products/802/509/314/0251/4.json' | ||
| server_domain = 'api.openfoodfacts.org' | ||
|
|
||
| expected_predictions_all = get_predictions_from_image(barcode, data.image, source_image, ocr_url) |
There was a problem hiding this comment.
We don't want to really call get_predictions_from_image because this would need to have the model, and we don't want to have it in dev / test environment.
That's why you will mock the function below.
Instead you should have something like:
expected_predictions_all = [
PredictionFactory(type=label, barcode=barcode, data={"model": "nutriscore", "confidence": 0.871872]}, value_tag="en:nutriscore-c", value="", automatic_processing="f")
PredictionFactory(type=label, barcode=barcode, data={"....Where parameter of your instances of PredictionFactory are taken from #115 (comment) but I don't take unnecessary values (some values are automatically generated by PredictionFactory, you can look at its code).
tests/data.py
Outdated
| @@ -0,0 +1,5 @@ | |||
| from robotoff.utils import get_image_from_url | |||
|
|
|||
| image = get_image_from_url('https://world.openfoodfacts.org/images/products/330/274/003/0949/front_fr.110.400.jpg', error_raise=False) | |||
There was a problem hiding this comment.
You misinterpret my previous command about putting the image in test/data, it was to put the image file in a data/ folder :-) (that is the jpg file).
But I think you don't need that.
tests/data.py
Outdated
|
|
||
| image = get_image_from_url('https://world.openfoodfacts.org/images/products/330/274/003/0949/front_fr.110.400.jpg', error_raise=False) | ||
|
|
||
| # expected_predictions_all = [Prediction(type=<PredictionType.nutrient: 'nutrient'>, data={'nutrients': {'salt': [{'raw': 'sel: 1,7 g', 'nutrient': 'salt', 'value': '1.7', 'unit': 'g'}]}, 'version': '2'}, value_tag=None, value=None, automatic_processing=None, predictor=None, barcode='3302740030949', timestamp=None, source_image='/802/509/314/0251/4.jpg', server_domain=None, id=None), Prediction(type=<PredictionType.nutrient_mention: 'nutrient_mention'>, data={'mentions': {'saturated_fat': [{'raw': 'acides gras saturés', 'span': [101, 120], 'languages': ['fr']}], 'fat': [{'raw': 'lipides', 'span': [87, 94], 'languages': ['fr']}], 'sugar': [{'raw': 'sucres', 'span': [151, 157], 'languages': ['fr']}], 'protein': [{'raw': 'proteines', 'span': [158, 167], 'languages': ['fr']}], 'salt': [{'raw': 'sel', 'span': [168, 171], 'languages': ['fr']}], 'nutrition_values': [{'raw': 'valeur nutritionnelle', 'span': [0, 21], 'languages': ['fr']}]}, 'version': '2'}, value_tag=None, value=None, automatic_processing=None, predictor=None, barcode='3302740030949', timestamp=None, source_image='/802/509/314/0251/4.jpg', server_domain=None, id=None), Prediction(type=<PredictionType.image_lang: 'image_lang'>, data={'count': {'fr': 25, 'null': 5, 'words': 35, 'en': 2, 'it': 3}, 'percent': {'fr': 71.42857142857143, 'null': 14.285714285714286, 'en': 5.714285714285714, 'it': 8.571428571428571}}, value_tag=None, value=None, automatic_processing=None, predictor=None, barcode='3302740030949', timestamp=None, source_image='/802/509/314/0251/4.jpg', server_domain=None, id=None), Prediction(type=<PredictionType.image_orientation: 'image_orientation'>, data={'count': {'up': 35}, 'orientation': 'up', 'rotation': 0}, value_tag=None, value=None, automatic_processing=None, predictor=None, barcode='3302740030949', timestamp=None, source_image='/802/509/314/0251/4.jpg', server_domain=None, id=None)] |
There was a problem hiding this comment.
I think that's what you get when you run the prediction locally, but as you don't have models for nutriscore locally (they are big models), this give you no prediction with value_tag="en:nutriscore-c". You only get predictions computed by other kind of predictors.
But as told below, this is not a problem, we are not interested in testing get_image_from_url but the import process.
|
|
||
| mock_get_predictions_from_image.assert_called_once_with( |
There was a problem hiding this comment.
Both assert_called_once_with throw the error saying assert_called_once_with shou d be called once but called 0 times
I was not able to resolve it. I did see the existing examples in the code. I read where to patch and tried to follow the steps in the document but no luck.
| "robotoff.insights.importer.import_insights", return_value=4 | ||
| ) | ||
| mock_get_predictions_from_image = mocker.patch( | ||
| "robotoff.insights.extraction.get_predictions_from_image" |
There was a problem hiding this comment.
Looking at the code I was not able to figure out what will get_predictions_from_image return
|
=========================== short test summary info ============================ |
|
Bug was closed. |
What
Why
The Nutri-Score model is not very reliable
Fixes bug(s)
Part of