medium · ~12 min
Product analytics logs feature activity sparsely in feature_usage (a row exists only when a user actually used a feature). The ML team needs a complete dense matrix covering every (user_id, feature) combination across the users and features dimension tables.
feature_usage
(user_id, feature)
users
features
users.user_id
features.feature
uses
0
user_id
feature
Input contract: user_id is unique and non-null in users; feature is unique and non-null in features; (user_id, feature) is unique and non-null in feature_usage, with non-null integer uses.