Additional features to the inference attack

[itrajanovska]

Hi,

Our team has a couple of suggestions which we think could be useful for other users as well.

1. In the inference attacks, it could be helpful to improve the attacks by using additional ML models beyond the provided k-NN model.
2. In order to enable within-group differences (to perform a fairness analysis), it could be beneficial to compute the risk at a group level, e.g different gender groups.

We implemented these two features in our forked version of anonymeter (https://github.com/itrajanovska/anonymeter).

Initially we implemented these features to support our package, Synthius (https://github.com/calgo-lab/Synthius) (https://pypi.org/project/synthius/).
However, since we plan to publish our package, we cannot keep GitHub repositories as direct dependencies (like our forked version of anonymeter).

Would you be open to including these extensions in the main repository, if you find them useful?
Alternatively, would it be acceptable for us to publish a forked version under a different name,
(with clear disclaimers that it is not the original package) and use it for our purposes?

We’re happy with whichever option works best for you and the project.

If you are open to including the changes to anonymeter, we'd be happy to open a PR.

[MatteoGiomi]

hey, thanks for reaching out! I'd be interested in including your contributions to this repo, so please feel free to open a PR. We can discuss the details there.

[itrajanovska]

Thanks for your response, here's the PR - #48