Computing individual privacy score (sMIA) for MIA Metrics Question

[jmp-3]

According to the Appendix F of the paper, in order to obtain the Aggregated Privacy Score, we need the individual privacy scores (sLOSS, sZLib, sMin-k, sMink++) for the metrics (LOSS, ZLib, Min-k, and Mink++).

I understand the harmonic mean aggregation is done post-hoc (already discussed in issue #150).

However, what is the transformation that I have to apply to the (LOSS, ZLib, Min-k, and Mink++) metrics to obtain (sLOSS, sZLib, sMin-k, sMink++) and match the Privacy Aggregate method of the original paper?

Thank you very much for the amazing work!