creme is a library for incremental learning. Incremental learning is a machine learning regime where the observations are made available one by one. It is also known as online learning, iterative learning, or sequential learning. This is in contrast to batch learning where all the data is processed at once. Incremental learning is desirable when the data is too big to fit in memory, or simply when it isn’t available all at once. creme’s API is heavily inspired from that of scikit-learn, enough so that users who are familiar with scikit-learn should feel right at home.

Most machine learning algorithms (be it supervised or unsupervised) assume a batch regime. However some of these algorithms have online variants. For example stochastic gradient descent is the online version of gradient descent whilst incremental k-means clustering is the online adaptation of k-means clustering. In general online algorithms perform slightly worse than their batch counterparts, although the gap is usually very small. However, online learning algorithms only consume a tiny amount of RAM, which thus makes them scalable and ideal candidates for commodity hardware and embedded systems. Moreover, they are much easier to put into production.

creme provides a nice interface for putting an incremental learning pipeline in place; a bit like what scikit-learn does for batch learning. Of course there are other open-source solutions available, but they are somewhat specialized towards certain tasks and can require a steep learning curve. Moreover some of these solutions aren’t “truly online” as they mostly assume the data is contained in a file. With creme it is possible to learn from a stream in it’s largest sense, be it a database query or a Kafka instance.