Install dependencies

Frontend

This is a create-react-app project.

To set up:

cd ./frontend
npm install
npm run build

This creates the static folder in the backend so that static files can be served by FastAPI. (Path is set in the .env file)

Backend

This is using Python3.

Create a virtualenv.

python -m venv venv 

Enter virtualenv (Windows).

venv/Scripts/activate

or (Linux)

source venv/bin/activate

Install requirements.

pip install -r requirements.txt

Running Locally

To run the API server:

uvicorn main:app --reload

The static pages can be accessed like this:

To test: http://localhost:8000/static/#product_code

e.g.

http://localhost:8000/static/#0677294998025

To run the frontend with dynamic reloading (in a new terminal):

cd ./frontend
npm start

To test: http://localhost:3000/#product_code

e.g.

http://localhost:3000/#0677294998025

Building the Docker Image

From the project root folder:

docker build --tag recipe-estimator .  

And to run:

docker run --name recipe_estimator -dp 5520:80 recipe-estimator

Processing Steps

Obtain Nutrients for Ingredients

For each ingredient we need to obtain the expected nutrient breakdown. This currently comes from the CIQUAL database, but other databases could be used, e.g. based on a regional preference.

If the ingredient on the product doesn't currently have a CIQUAL code then attempt to look this up is made using ingredients.json. To refresh the ingredients taxonomy you can use the followng script. I have currently formatted the JSON before committing:

curl https://static.openfoodfacts.org/data/taxonomies/ingredients.json --output ciqual/ingredients.json

Having found the ingredient in CIQUAL a nutrient map is added to each ingredient. Only the "main" nutrient is used (one without an underscore suffix).

Determine nutrients for computation

Only nutrients that occur on every ingredient can be used. Energy is also eliminated as this combines multiple nutrients.

Weighting

A weighting can be applied to give specific nutrients more / less impact on the overall calculation. If a nutrient has no weighting then 1 is assumed.

Estimation

The estimation attempts to find the proportion of each ingredient that minimises the weighted difference between the computed nurtients and those obtained from the product.

Return data

An example return structure is shown below:

ingredients: [
  {
    id: "en:tomato",
    percent_estimate: 67.2,
    evaporation: 4,
    nutrients: {
      calcium: {percent_min: 0.023, percent_max: 0.024},
      carbohydrates: {percent_min: 3.45, percent_max: 3.45},
      ...
    }
  },
  ...
],
recipe_estimator: {
  nutrients: {
    calcium: {
      product_total: 0.06,
      weighting: 1000,
    },
    vitamin-b2: {
      notes: "Not listed on product"
    },
    ...
  },
  ingredient_count: 3,
  iterations: 35,
  status: 0,
  time: 0.2
}

Ingredients

The original ingredients map will be returned with additional percent_estimate field and a nutrients map with the expected nutrient value ranges in g per 100 g/ml of that ingredient.

A quantity_estimate field is also provided which shows the amount of the ingredient needed to make 100 g/ml of the product. Note this may be higher than the percent_estimate because of evaporation during the product preparation / processing. An evaporation field shows the estimated water loss during processing.

Recipe Estimator

A new "recipe_estimator" map will also be returned, providing the compted nutrients and some metrics

Nutrients

This will contain the calculated value based on the new estimate and will also provide the weighting used, the nutrient identifier for the database used (CIQUAL), the quoted proportion from the product and the difference from the computed value. Nutrients that were not quoted on the product will also be included.

Metrics

This will provide details of the computation performed, such as time taken and number of iterations.

TODO

• Need to return a more formal error object
• Use min and max for ingredient nutrient when stated as "< ..." in CIQUAL
• Use min and max from CIQUAL for unmatched ingredients
• Cope with min and max on product nutrients (e.g. if we had to get from a category)

Background Info

To get nutrient types for nutrient_map.csv I used:

db.products.aggregate([
  {
    $project: {
      keys: {
        $map: {
          input: {
            "$objectToArray": "$nutriments"
          },
          in: "$$this.k"
        }
      }
    }
  },
  {
    $unwind: "$keys"
  },
  {
    $group: {
      _id: "$keys",
      count: {
        "$sum": 1
      }
    }
  }
])

Need to skip any nutrients where Ciqual value is '-' as this means not known, not zero