Changelog#

This is a record of all past maite releases and what went into them, in reverse chronological order.

0.7.3 - 2025-02-19#

  • Update GitHub CI/CD to not directly publish docs (which is now done through GitLab mirroring)

  • Update GitHub CI/CD to use latest version of upload-artifact and download-artifact actions

0.7.2 - 2025-02-12#

  • Add how-to guide for wrapping object detection models & datasets

  • Add how-to guide for wrapping image classification models & datasets

  • Add docstring examples for object detection model, dataset, metric, & augmentation

  • Add docstring examples for image classification model and metric

  • Add image classification tutorial

  • Back MAITE ArrayLike with numpy.typing.ArrayLike

  • Publish historical and latest built documentation to github (automatically)

0.7.1 - 2024-11-06#

  • bugfix: correct typo github action for publishing new docs to gh-pages

0.7.0 - 2024-11-06#

  • Add TypedDict metadata attributes to Dataset, Augmentation, Model, Metric protocols

  • Add support for Python 3.11 and drop support for Python 3.8

  • Automatically generate documentation from notebooks in examples

  • Display progress-bar when maite Dataset is passed to evaluate

  • Automatically typecheck docstring examples

0.6.1 - 2024-10-16#

  • bugfix: Added upper limit to numpy dependency ("<2") for compatibility with torchvision

  • Updated file headers/copyrights

  • Added new ci job for testing pypi publishing

  • Update image references in torchvision_object_detection tutorial

0.6.0 - 2024-06-14#

  • Removed torch dependency (internally, input/target types are batched agnostically of task)

  • Altered batch types in each task to be sequences of singleton types

  • Added tqdm progress bars to evaluate/predict workflows

  • Added docstring examples to Dataset and Augmentation protocols

0.5.0 - 2024-03-28#

This is a large change to protocols based on experiences of our growing userbase. These updated protocols should enable a wider variety of use cases, but do contain backward-compatibility breaking changes.

High-level overview:

More details:

  • Class structure of core component protocols is now ML subproblem agnostic – e.g. Model objects must all implement the same named methods.

  • Type signatures of core component protocols are specific to ML subproblem domain, permitting subproblem specific component implementers to know more about their expected inputs/outputs.

  • Within an ML subproblem, the model input type, model target type, datum metadata type (and the 3 respective batched versions of these types) are prescribed.

  • Core protocols (with the exception of Dataset) are only required to handle data in batched form (this may change)

  • evaluate and predict functions now exist under maite.workflows to separate them from core protocols

  • evaluate function takes ML components from either image processing or object detection domain (component domain compatibility enforced statically)

  • predict function permits running model inference without evaluate

  • “Interop” functionality (i.e. ability to load maite-wrapped versions of components originating from some third parties) has been rolled back temporarily as core architecture changed.

0.4.0 - 2024-01-22#

  • bugfix: update maite exception naming and fix typo in import

  • feature: added GitHub workflow to build and publish public documentation

  • feature: added the use of DatumMetadata in tutorial basic_evaluation.ipynb

0.3.6 - 2024-01-17#

This release includes the following changes:

  • Fix issue with loading datasets from huggingface hub (e.g., cifar10) (bugfix)

  • Import dataclass from dataclasses instead of attr (bugfix)

  • Remove version constraints for torchmetrics

0.3.5 - 2024-01-05#

This release includes two new features:

  • Add dataset and datum level metadata

  • Add model metadata

0.3.4 - 2023-12-21#

Publish to pypi.org

0.3.3 - 2023-12-21#

Modifying pyproject.toml to obtain better author display representation on pypi

0.3.2 - 2023-12-21#

Updated README and pyproject.toml in preparation for pypi publishing (Note: we are still publishing to test.pypi)

0.3.1 - 2023-12-21#

Updated docs

  • Updated references in docs/README.md and docs/index.rst to reflect current repo name

    • Updated basic_evaluation.md tutorial

    • Implemented default parameters for object detection

  • Added .github/pypi_publish.yml directory to automate publishing the repository to online packaging indices (after GitLab is mirrored to GitHub)

0.3.0 - 2023-12-20#

This release is the first release after renaming to maite, the changes are follows

  • The major change is the rename jatic-toolbox to maite

  • It adds headers (copyright, license) to each file

  • It adds the Phase-1 protocol and base provider/hub registration system from Quansight.

0.2.0 - 2023-07-28#

This release provides a number of updates for usability and stability.

Protocols#

Many updates to the names of protocols, usability, and stability. See reference documentation: https://jatic.pages.jatic.net/cdao/maite/api_reference.html

For an explanation of usage of current protocols see (deprecated link) explanation/protocols_current.

For an overview of the future vision of protocols see (deprecated link) explanation/protocols_vision.

Documentation#

Explanations, how-tos, and tutorials have been added to the maite documentation at https://jatic.pages.jatic.net/cdao/maite/.

Additionally, be sure to check reference documentation for examples of usage and API details. See https://jatic.pages.jatic.net/cdao/maite/api_reference.html.

Model Inference#

Previous release supported dictionary inputs to models. This release changes the API to support to a single or collection of arrays. This is more consistent with model inference APIs utilized in most deep learning frameworks.

  • See image classification model definition: maite.protocols.ImageClassifier

  • See object detection model definition: maite.protocols.ObjectDetector

Additionally, the use of pre-processors and post-processors is kept internal to model inference rather than explicitly requiring users to manage these steps. Integration with augmentations and perturbations is still in development.

Stability and Usability#

  • Remove 3.7 support and update typing_extensions references

  • Pins torchmetrics < 1.0 due to breaking API changes

  • Number of bug fixes and stability improvements

  • Improved testing to ensure toolbox protocols and testing utilities work with minimal installation.

  • Improve testing coverage for dataset and model loading

  • Initial implementation of dataset, model, and metric registries. See (deprecated link) how_to/named_evaluation.

0.1.0 - 2023-05-12#

This marks the first release of the maite. We are not yet at a stable v1.0.0. Future release will aim to improving testing and stability of the software for general use.