Closed ddahlbom closed 8 months ago
Overall looks good to me, but maybe one test got weakened. Eventually we might want to consider merging with an OnlineStatistics
structure, but that's not available yet.
Sorry, the weakened test was an oversight. It is fixed.
Replace running calculation of variance with running calculation of
M=(n-1)*σ²
, wheren
is the number of samples.The modifications reproduce previous results when
M
is divided byn-1
. Note that there is currently no interface for retrieving this data. Any future interface will need to divide byn-1
to recover the variance.