vbursztyn / recipe-generator Goto Github PK

Based on [1][2][3], the Nutrient Rich Foods Index 9.3 (NRF 9.3) could come in handy:
(i) Three bad ingredients - find whether saturated fat, sugar, or sodium can be minimized;
(ii) Nine good ingredients - find whether protein, fiber, vitamins (A, C & E), and minerals (Ca, Fe, Mg & K) can be maximized.

*See method fetch_nutrition_facts() in RecipeFetcher for building the diagnosis part.

[1] Hess, Julie M., and Joanne L. Slavin. "Healthy snacks: using nutrient profiling to evaluate the nutrient‐density of common snacks in the United States." Journal of food science 82.9 (2017): 2213-2220.
[2] Drewnowski, Adam. "The Nutrient Rich Foods Index helps to identify healthy, affordable foods." The American journal of clinical nutrition 91.4 (2010): 1095S-1101S.
[3] https://depts.washington.edu/uwcphn/news/presentations/NRF_041609.pdf

After diagnosing the problems with a particular recipe (see issue #3), we can suggest a treatment (ingredient replacement) following certain criteria (e.g., NRF 9.3). To do so, we may need to rank possible replacements by their nutritional gains; thus, we may need to tap into an ingredient-centered nutritional database. We can do this by using python-usda [1] to access the National Nutrient Database for Standard Reference [2].

Suggestion: after pairing ingredients in 'ingredient-ingredient_network', we can add nutrients to network nodes.

[1] https://pypi.org/project/python-usda/
[2] https://ndb.nal.usda.gov/ndb/foods/show/11333

Implement method parse() in class RecipeFetcher, or equivalent.

Generation of list:

1. Base DB is FooDB [ http://foodb.ca/ ]

2. Selected subcategories:

• protein_list = ['Fishes','Swine','Bovines','Crustaceans','Poultry','Animal fats','Animal foods','Meat products','Ground meat','Fish Products','Other mammals','Animal fats','Amphibians','Venison','Mollusks','Marsupials','Rodents','Cetaceans','Equines']

• condiment_list = ['Seasonings','Dressings','Sauces','Condiments','Herbs and Spices','Spices','Herbs','Nuts']

3. Data generation:

• df_foods[df_foods['food_subgroup'].isin(protein_list)][['name','food_subgroup']].to_csv(...) -- 193 tuples

• df_foods[df_foods['food_subgroup'].isin(condiment_list)][['name','food_subgroup']].to_csv(...) -- 121 tuples

1. Search keywords related to each international cuisine to assemble a list of top recipes;
2. Fetch each one of said recipes and retrieve the associated ingredients;
3. Identify (i) the most frequent ingredients and (ii) the most frequent ingredient pairings within each set.

Suggestions: start with Italian, French, Mexican, Indian, and Japanese cuisines. Consider expanding it after getting to a MVP.

Just saw a recipe with the following

STATEMENT: 1 (1.5 fluid ounce) jigger Irish cream liqueur
NAME: (1.5 fluid ounce) jigger Irish cream liqueur
BASETYPE: None
QUANTITY: 1.0
QUANTITY MODIFIER: None
UNIT: None
PREP STEPS: []

Pick (i) a viewpoint towards ingredients, (ii) a network output format, and create a class to generate a network with ingredient pairings (i.e., possible replacements).
Suggested high-level contract: place raw ingredients data in folder 'ingredients_data' and the generated network in folder 'ingredient-ingredient_network.'
Suggestions for (i): cooccurrence in reviews [1] or shared flavor compounds [2].
Suggestions for (ii): for each ingredient, store ingredients at one hop and two hops (as mongoDB collections or directly in the file system).

[1] Teng, Chun-Yuen, Yu-Ru Lin, and Lada A. Adamic. "Recipe recommendation using ingredient networks." Proceedings of the 4th Annual ACM Web Science Conference. ACM, 2012.
[2] https://github.com/lingcheng99/Flavor-Network