Thetis 0.2.5

Thetis 0.2.5 Latest

We're happy to announce that Thetis version 0.2.5 is available.

The key changes of version 0.2.5 are:

• Integration of a system description argument in the public interface and documentation
• Alignment of thetis-public and thetiscore interfaces
• Improvements to evaluation pipeline structure and modularity
• Dependency cleanup and removal of unused components
• Minor bug fixes and interface improvements

The public interface has been extended to include an AI system description to support technical documentation requirements such as those defined in the AI Act (Annex IV).

Thetis 0.2.4

Thetis 0.2.4 Latest

We're happy to announce that Thetis version 0.2.4 is available.

The key changes of version 0.2.4 are:

• Update Documentation
• Update detection.ipynb example
• Bugfixes

API key as surrogate for license file which can be obtained on https://app.thetis.de/ .
Include the AI system description according to the technical documentation requirements of the AI Act (Annex IV).
Some minor bugfixes.
Update of example/detection.ipynb.

Thetis 0.2.3

We're happy to announce that Thetis version 0.2.3 is available. We now support API keys and include the AI system description according to the technical documentation requirements of the AI Act (Annex IV).

The key changes of version 0.2.3 are:

• API key as surrogate for license file which can be obtained on https://app.thetis.de/
• Include the AI system description according to the technical documentation requirements of the AI Act (Annex IV)
• Some minor bugfixes

Thetis 0.2.2 with new freemium mode

We're happy to announce that Thetis version 0.2.2 is available. One of the key changes is that Thetis now supports a freemium mode, allowing to integrate Thetis in your current workflow without requiring a valid license.

The key changes of version 0.2.2 are:

• Freemium variant with free performance evaluation. For the packages "fairness", "uncertainty", and "data_evaluation", a valid license is still required.
• Make report name configurable.
• Enhanced diagram serialization.
• Minor bugfixes.

Thetis 0.2.1

Release for Thetis 0.2.1 with the following bugfixes:

• Fixed bug where requirements in thetiscore where missing.
• Fixed bug where wrong dataset information was provided for detection example.

Thetis 0.2.0

We're happy to announce that Thetis v0.2.0 is now released with support for regression tasks and an updated user experience!

Release notes:

• Enable complete regression safety evaluation.
• Update structure of configuration file.
• Update and fine-tuning of recommendation system.
• Make bounding-box clipping user configurable.
• Make bounding-box merging strategy user configurable (exclusive (default) or max).
• Rework performance evaluation in classification to use sensitivitiy, specificity, precision, NPV, informedness, markedness, and MCC.
• Rework fairness evaluation in classification that is now based on the ratios of metrics provided by the updated performance module.
• Minor bugfixes.

Thetis 0.1.4

The latest Thetis release contains only small bugfixes.

Update notes:

• Fix a bug where the data set rating might have exceed the [0, 10] scale.

Thetis 0.1.3

Today we deliver the next minor update for our Thetis alpha version. This update contains several minor bugfixes and provides further application examples as Jupyter notebooks. Furthermore, we provide additional helper functions for relevant I/O operations to store/reload AI model predictions in the right format for Thetis.

Update notes:

• Provide new io package with functions read_json_with_pandas and write_json_with_pandas to load/store data in the right Thetis input format.
• Provide Jupyter notebooks for advanced application examples.
• Provide test application licenses for the Jupyter notebook examples.
• Minor documentation updates.
• Minor bugfixes in Thetis core.

Thetis 0.1.2

We are happy to announce Thetis 0.1.2 open alpha version!
Thetis is a toolbox to automatically examine, evaluate and rate your AI-based system w.r.t. several safety aspects, including:

• Fairness
• Uncertainty consistency (calibration)
• Data set quality
• Baseline AI performance metrics

You can simply install Thetis by

pip install thetis

Hint: you need a valid license to run Thetis. Please visit https://www.efs-techhub.com/efs-portfolio/loesungen/thetis to obtain one.
For a detailed documentation, please visit https://efs-opensource.github.io/Thetis/index.html.