Discussed minimizing the trusted codebase with Adam. The idea is that we could only have a single function for clipping, data imputation, etc. that only operates on single dimensional vectors, then just use a wrapper function for matrices and more complex objects.
If the wrapper for this is external to the mechanisms (ie as part of the dpStatistic$release function), which would minimize the trusted codebase the most, this will potentially increase inefficiencies in things like the covariance statistic, as it means the individual covariances in the covariance matrix will need to individually call the Laplace mechanism.
Discussed minimizing the trusted codebase with Adam. The idea is that we could only have a single function for clipping, data imputation, etc. that only operates on single dimensional vectors, then just use a wrapper function for matrices and more complex objects.
If the wrapper for this is external to the mechanisms (ie as part of the dpStatistic$release function), which would minimize the trusted codebase the most, this will potentially increase inefficiencies in things like the covariance statistic, as it means the individual covariances in the covariance matrix will need to individually call the Laplace mechanism.