High-Level Tasks

1. BJCP vs. Empirical

Exploratory data analysis Compare the BJCP and real recipes:

(ABV, IBU, colour)
Characteristic ingredients

Deliverable? Interactive plots / tables

Lookup table of style -> ranges of properties.
IBU vs ABV (Rory likes this)
- Color of point represents SRM.
- Size of point could represent number of recipes or spread of data.
- Size could also be spread of data
- Could also do contour lines
- FG, OG
- Matrix plots - can show relationships of all 4 variables. Highlighting section of data in 1 plot shows same points in other relationships
Word cloud
- Beer names
- Hop varieties
- Styles?
Hop vs boil time (Rory also likes this)
- Show amount of hop added at each point in boil time for different beers
- Compare style A to style B
- Show how this changes over time

2. Generate recipe for style X

For a given style, what is a "typical" recipe? Generate a vector in ingredient-space.

3. Generate recipe interpolating between styles X, Y

For a given set of styles, generate a recipe that captures the defining characteristics of both styles.

4. Where does a recipe sit in its style?

For a given style and recipe, how "typical" is the recipe?

5. Generate random recipe

Generate a recipe that is:

Uncommon
Not terrible
- Identify patterns in ingredients that produce good flavours
- Which are still uncommon

Low-Level Tasks

Website

Add a collection-type page for apps
Decide on a list of posts
Figure out how to post a Dash/Plotly.js app

Documentation

Add tabular data dictionary for fields in all\_recipes.h5
- Core
- Ingredients
Add super high level overview of data pipeline
- .xml ---(converter.py)---> .h5 ---(recipe2vec)---> .h5