We need to identify how to properly update all micro-clusters if renormalizing: I believe we can simply multiply the means by a ratio where <1 indicates a new maximum was observed, and >1 indicates a new minimum was observed, but variance is a different story. We cannot multiply the variance (or the sum of the squares) by such a ratio and get a correct result.
We need to identify how to properly update all micro-clusters if renormalizing: I believe we can simply multiply the means by a ratio where <1 indicates a new maximum was observed, and >1 indicates a new minimum was observed, but variance is a different story. We cannot multiply the variance (or the sum of the squares) by such a ratio and get a correct result.