This code is inspired by fastai, but differs in implementation in several ways. Most importantly, the original Python code makes heavy use of side-effects where the Learner
holds different state variables, and other objects access and modify them.
This has been replaced by a more functional design. The state is now transmitted via arguments to Callbacks
which may then pass them on to Metrics
.
Note: this is a package in-development. Expect breaking changes for the foreseeable future, but we want you to test out the package by following the documentation. Any contributions are welcome via PRs/issues.
Much of the documentation has been copied from the original Python, and modified where appropriate.