Recipe/Research ideas
Listing some ideas to explore
Variance distribution
The fitting procedure based on non-negative least squares with a forced order very often gives an excellent fit. But sometimes the fit does not work at all. We need better understanding when what happens.
A better understanding is needed for the active and non-active ingredients. Which ones are more active than others?
- What is the best cutoff variance to decide between good and bad recipe estimations? A variance of 1.0 seems reasonable, due to the quadratic nature of the variance.
- What are the reasons behind bad recipe estimations? Ingredient variability, producer error, ...
- Does the estimation quality depend on the number of ingredients? The more ingredients, the easier to find the recipe?
- Are there ingredients that are more often activated than others? Are the single nutrient ingredients?
- What are the ingredients that are most deactivated? With a larger water fraction?
For all of these questions we need a large set of products that have been fitted.
Ciqual alternatives
Are there other ingredient databases that could be used? This is also needed for industrial ingredients and more exotic ones.
OFF ingredients
Can the OFF ingredients data be used? mean/deviation/raw data/fitted data
Missing ingredients
How to find the ingredient data that is missing in Ciqual, like cocoa paste.
Category recipe
Can we use the data to define a recipe for a category?
Function recipe
Is it possible to define a recipe by using distribution fits?
Nutrients profile taxonomy
Some ingredients have exactly the same nutrients, think of sugars or oils. This makes it impossible for a recipe estimator to distinghuish between say colza oil or sunflower oil. You can only say that the recipe contains oil. Oil can then be seen as a parent ingredient of sunflower oil. All children of this oil will have the same nutrients. Are there more cases like this? Is this relevant?