Getting started¶
Examples¶
Currently the best place to start is the example jupyter notebooks. You can download them or run them on Google Colab.
Please open an issue on GitHub if you have any questions.
Overview¶
This library consists of 12 modules:
| Module | Description |
|---|---|
| Adapters | Wrappers for training and inference steps |
| Containers | Dictionaries for simplifying object creation |
| Datasets | Commonly used datasets and tools for domain adaptation |
| Frameworks | Wrappers for training/testing pipelines |
| Hooks | Modular building blocks for domain adaptation algorithms |
| Inference | Algorithm-specific functions used during inference |
| Layers | Loss functions and helper layers |
| Meta Validators | Post-processing of metrics, for hyperparameter optimization |
| Models | Architectures used for benchmarking and in examples |
| Utils | Various tools |
| Validators | Metrics for determining and estimating accuracy |
| Weighters | Functions for weighting losses |
Adapters¶
Adapters contain an algorithm's training step and inference step. The training step is defined in the wrapped hook.
View this notebook for examples.
Containers¶
Containers are Python dictionaries with extra functions that simplify object creation.
View this notebook for examples.
Datasets¶
The datasets module consists of wrapper classes that output data in a format compatible with hooks.
It also contains some common domain-adaptation datasets like MNISTM, Office31, and OfficeHome.
View this notebook for examples.
Frameworks¶
This library works with plain PyTorch, but what about PyTorch frameworks like Ignite, Lightning, and Catalyst?
Of course, anything that works in PyTorch will also work in PyTorch frameworks. However, there is some setup involved, like registering event handlers in Ignite, or writing class definitions in Lightning and Catalyst.
The purpose of the frameworks module is to eliminate that setup.
Check out these notebooks for examples:
Hooks¶
Hooks are the building blocks of the algorithms in this library.
Check out these notebooks for examples:
Layers¶
The layers module contains loss functions and wrapper models. These are used in combination with hooks to create domain adaptation algorithms.
Validators¶
Validators compute or estimate the accuracy of a model.
Weighters¶
Weighters multiply losses by scalar values, and then reduce the losses to a single value on which you call .backward().