TODO:
- take the biz categories from the user biz interactions and create normalized feature vector per user re-write or write new user vector. look for trending businesses with categories matching to user
- recommendation strategy, collaborative filtering
- refactor unit & itests
- get spark runs twice
- split out tests
- make parquet reader part of itest
ML ideas:
- predict if rfn based on review text
- predict star rating
Search ideas:
- inverted index
- load into es
Data ideas:
- test joining different column groups
- find a dataset with city, state lat lon append to each business
- one hot encode business categories ie, has_restaurants
create a time dimension, run for year x (and data < x)
Steps for adding a column group:
- try joining with existing df in an interactive session
- print out schema, save it in schemas directory
- regenerate python client to join all column groups
- test client