f-DP, posterior risk and relative risk

[damienbfs]

relative risk.

[Shoeboxam]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• f-DP, posterior risk and relative risk #1816 : 2 dependent PRs (#1828 , #1868 ) 👈
• add make_fixed_approxDP_to_approxDP #1860
• Replace ε(δ) curve with δ(ε) curve #1831 : 1 other dependent PR (#2001 )
• rename SMDCurve to PrivacyProfile #1999
• Approximate<M> privacy measure #1768 : 2 other dependent PRs (#1854 , #2008 )
• remove generics from measures #2007
• Fix: controlled rounding for sub-sub-normals #1998
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @damienbfs and the rest of your teammates on Graphite

[Shoeboxam]

Not a full review, but while writing FFI I noticed that creating a tradeoff curve with num_approximations=0 panics. In general, try to avoid unwrapping and panicking in Rust, because it will cause the Python or R interpreter to shut down.

In cases where you want to quickly return an error, and have an error, look into the try operator in rust (?).

[Shoeboxam]

Also, given that this is another form of a privacy guarantee from the library, the final result should be computed with conservative arithmetic. If you don't want this to affect compute performance of the function, you can conduct the search in standard arithmetic and then only compute the conservative version once (as done in the δ(ε) rust conversion function I sent you).