This project was a big challenge. Had various steps, various discoveries, various steps. But, ultimately, had one goal: Create a basis for a Meal Planner or Recipe Recommender for the average person that doesn't have time to think about food everyday.
It started with scraping Recipes from a magazine with almost 50 years and lots of people messing with Data on the website. After the scraping, the cleaning of Data, the decisions of what and how to store the Data I gathered, it was the decision of how to use it. I decided to try Clustering and Cosine Similarity. Cosine Similarity won, even if not because Similarity had a ring to it. Similar recipes to the ones the user asked for. The Main Program works. The App in Streamlit is a work in development. The Natural Language Processing I began on Preparations gave me ideas.
Lots of Improvements. This was above all a project to learn. To produce mistakes, to understand that Data is immense and must be treated carefully. Cosine Similarity was a mistake. Probably Clustering, Random Forests would be better. Considering that the Recipes Features are a mix of Categorical and Numerical Variables, it should have been a mix. In the end it works. But, it is definitely a project in the making for the foreseable future with the adding and repairing of some features like Ingredients or the Pricing. To do Servings per Person and their respective pricing.