ingredient-parser-nlp 2.5.0

A Python package to parse structured information from recipe ingredient sentences

Ingredient Parser

The Ingredient Parser package is a Python package for parsing structured information out of recipe ingredient sentences.

Documentation

Documentation on using the package and training the model can be found at https://ingredient-parser.readthedocs.io/.

Quick Start

Install the package using pip

$ python -m pip install ingredient-parser-nlp

Import the parse_ingredient function and pass it an ingredient sentence.

>>> from ingredient_parser import parse_ingredient
>>> parse_ingredient("3 pounds pork shoulder, cut into 2-inch chunks")
ParsedIngredient(
    name=[IngredientText(text='pork shoulder', confidence=0.996867, starting_index=2)],
    size=None,
    amount=[IngredientAmount(quantity=Fraction(3, 1),
                             quantity_max=Fraction(3, 1),
                             unit=<Unit('pound')>,
                             text='3 pounds',
                             confidence=0.999982,
                             starting_index=0,
                             unit_system=<UnitSystem.US_CUSTOMARY: 'us_customary'>,
                             APPROXIMATE=False,
                             SINGULAR=False,
                             RANGE=False,
                             MULTIPLIER=False,
                             PREPARED_INGREDIENT=False)],
	preparation=IngredientText(text='cut into 2 inch chunks',
                               confidence=0.999946,
                               starting_index=5),
	comment=None,
	purpose=None,
	foundation_foods=[],
	sentence='3 pounds pork shoulder, cut into 2-inch chunks'
)

Refer to the documentation here for the optional parameters that can be used with parse_ingredient .

Model

The core of the library is a sequence labelling model that is used to label each token in the sentence with the part of the sentence it belongs to. A data set of 81,000 example sentences is used to train and evaluate the model. See the Explanation section of the documentation for more details.

The model has the following accuracy on a test data set of 20% of the total data used:

╒══════════════════════════╤══════════════════════════╕
│ Sentence-level results   │ Word-level results       │
╞══════════════════════════╪══════════════════════════╡
│ Accuracy: 95.25%         │ Accuracy: 98.09%         │
│                          │ Precision (micro) 98.07% │
│                          │ Recall (micro) 98.09%    │
│                          │ F1 score (micro) 98.08%  │
╘══════════════════════════╧══════════════════════════╛

Development

Basic

Train and fine-tune new ingredient datasets to expand beyond the existing trained model provided in the library. The development dependencies are in the requirements-dev.txt file. Details on the training process can be found in the Explanation documentation.

Web App

The ingredient parser library provides a convenient web interface that you can run locally to access most of the library's functionality, including using the parser, browsing the database, labelling entries, and training the model(s). View the specific README in webtools for a detailed overview.

Parser	Labeller	Trainer

Documentation

The dependencies for building the documentation are in the requirements-doc.txt file.

Contribution

Please target the develop branch for pull requests. The main branch is used for stable releases and hotfixes only.

Before committing anything, install pre-commit and run the following to install the hooks:

$ pre-commit install

Pre-commit hooks cover both the main python library code and the web app (webtools) code.