We can start by asking each time the DP-size of the dataset for the weighted aggregation (and cut a part of the privacy budget accordingly), like for the mean function. In a second step (independent of this issue), we will try to cache the DP-size to avoid spending budget on this).
For tricky methods like variance / covariance, we will need further implementations, so let's put this on a different issue for later.
We can start by asking each time the DP-size of the dataset for the weighted aggregation (and cut a part of the privacy budget accordingly), like for the mean function. In a second step (independent of this issue), we will try to cache the DP-size to avoid spending budget on this).
For tricky methods like variance / covariance, we will need further implementations, so let's put this on a different issue for later.