v0.3.4
Important
This version includes a substantial new submodule, abstract_dataloader.ext.
What's Changed: Added several ADL "extension" modules which are not imported by default. These extensions provide new functionality that extends the core spec, but are not necessarily universally applicable (but should still be broadly applicable to most settings).
- augment: A protocol for specifying data augmentations.
- graph: A programming model based on composing a DAG of callables into a single transform.
- lightning: A lightning datamodule wrapper for ADL datasets and pipelines.
- objective: Standardized learning objectives, and a programming model for multi-objective learning.
- sample: Dataset sampling utilities, including a low-discrepancy subset sampler.
- types: Type-related utilities which are not part of the core ADL spec.
Full Changelog: v0.3.3...v0.3.4