Back to Projects

Build from first principles

Evaluate a Recommender Honestly

Write Recall@K, NDCG and coverage, split by time, and see how a random split makes a weak model look great.

How do you know a recommender works?See what offline evaluation measures and where it goes wrong
From Metrics to an Honest TestSee how your functions fit together before the notebook
Test Five Recommenders on Real OrdersCompare a random split with a time split and find the leak