Case study
Grocery Purchase Analysis
A recipe suggestion analysis over a dataset of recipe ingredients, plus purchasing statistics by retailer, built as the first stage of a larger application.
Overview
The analysis half of a recipe application. Given a dataset of ingredients per recipe, it works out what can be cooked from a chosen set of ingredients and plots the statistics behind the suggestions.
One of the analyses produced under the data-driven content contract.
What I own
The analysis and the visualisations.
Complexities tackled
Suggesting recipes is a set problem, and the interesting part is the near misses. Recipes that can be made exactly are the trivial answer. The useful answer includes recipes that are one or two ingredients short, ranked by how close they are, because that is what actually changes what someone buys.
Ingredient names are not identifiers. The same thing appears under several names, in different units and at different levels of specificity, and none of the matching works until that is normalised.
Analysis on the fly, not precomputed. The suggestions respond to a chosen set rather than to a fixed query, so the work has to be cheap enough to run per request rather than baked ahead of time.
Stack
Python with Pandas for the analysis, Matplotlib for the charts.
Working on something similar?
Tell me what you are building and what is in the way. I will tell you honestly whether I am the right person for it.