# Deliveroo
# How Deliveroo improved the ranking of restaurants
# About
Company is 3 sided marketplace.
10k restaurants 10k riders consumers in 300 cities across 14 countries.
Given a list of restaurants , rank them optimally.
# Algorithm
Treat is as a pointwise classification problem Probability of user purchasing from the restaurant
# Metrics:
Offline: Mean Reciprocal Rank, Precision at k
Online: Order Volume
Session Level conversion = # sessions that resulted in an order / # sessions
# Productionization
Use Go with tensorflow
Employees can look at their individual restuarant lists with explanations
Look at Spearman Rank correlation for different ranking algorithms.
# Future
Looking into more features and wide and deep models
