Release 7

Added the colourised MNIST dataset. In this dataset during training the colour of the digit is a shortcut to the class label, but at test time, this is not the case.

Release 6

Introducing EthicML Vision.

We've found that we're using image datasets more and more. Fitting in with the goals of EthicML, we wanted to abstract away all the boring stuff so that it's easy to get going with a dataset which accounts for a sensitive label in addition to the features and class label.

Separation of async

We've also made the decision to make some (rather than all) methods run in their own process. These typically are either long running, or can be accelerated with a gpu. "Quick Running" processes no longer happen asynchronously.

Release 5

Minor updates

Agarwal uses fairlearn's latest release.
Typed Arg Parsing
Minor updates to typing library we use

EthicML 0.1.0-alpha3

Changes

• Add Zafar's algorithm
• Add plotting function for mean-std box plots
• Improve evaluate_models() massively
• Add fair grid search
• Add preliminary version of Hardt's algorithm
• Internal improvements

EthicML 0.1.0-alpha2

Changes

• Made the Datatuple and Testtuple classes iterable
• Added Upsampler pre-processing
• Added CORELS algorithm for the ProPublica data
• Improve documentation (thanks @MylesBartlett )
• InstalledModel can now be configured to not download things
• General bugfixes

We had a think

and decided to make the name property of DataTuples optional. The DataTuple object is really useful for being able to store your data in an ad-hoc way, so we want to make sure that until there's a good reason not to, that this can remain part of your workflow.

Hello World!

This is our first release for EthicML - a researcher's toolbox for benchmarking fair ML models.

We hope that it's useful to you, but despite being on-par with other fair ml toolkits it still has a loooong way to go. So, if you come across an issue, or have a feature request, please either raise an issue or open a PR.

✌️