1: Averages have their limitations, especially for my needs, as they tend to smooth out significant differences between samples. In my case, the peaks and valleys in the data were crucial, so the 3buckets algorithm was developed to preserve these inflection points while disregarding repetitive or similar data.
For instance, when dealing with temperature data that remained relatively constant until a certain point (sample X) where even a slight percentage change became significant, the buckets algorithm proved to be ideal.
Moreover, the flexibility of the parameterized buckets algorithm allows for adjustments in values to determine the threshold at which excessive data removal begins to compromise visual representation. This process enables users to fine-tune the amount of data retained in the graph.
Instead of a simplistic decimation approach (removing X out of N samples), wherein outliers may inadvertently be eliminated, this algorithm offers a more sophisticated alternative. By implementing criteria based on the magnitude of difference, outlier samples can be safeguarded from being filtered out unnecessarily.
2: The feasibility of computation, data dynamics, and other variables influence the choice of algorithm. In my experience, once data is archived and a representative sample is selected for each bucket, there's no need for recalculations—thus ensuring consistency and efficiency.
Given the age of your query, I'm curious to know what solution you eventually implemented?