
AI Blueprints
By :

We will build a simple recommendation system that may be easily integrated into an existing platform. Our recommendation system will be deployed as an isolated HTTP API with its own internal memory of purchases (or clicks, or listens, and so on), which is periodically saved to disk. For simplicity, we will not use a database in our code. Our API will offer recommendations for a particular user and recommendations for similar items. It will also keep track of its accuracy, explained further in the Continuous evaluation section.
The bulk of the features of our recommendation system are provided by Ben Frederickson's implicit
library (https://github.com/benfred/implicit), named as such because it computes recommendations from implicit feedback. The library supports the ALS algorithm for computing the matrix factorization described previously. It can use an internal nearest neighbor search or faiss
(https://github.com/facebookresearch/faiss...