On Wed, Jan 18, 2023 at 1:02 PM Peter Geoghegan <pg@bowt.ie> wrote:
> Some of what I'm proposing arguably amounts to deliberately adding a
> bias. But that's not an unreasonable thing in itself. I think of it as
> related to the bias-variance tradeoff, which is a concept that comes
> up a lot in machine learning and statistical inference.
To be clear, I was thinking of unreservedly trusting what
pgstat_report_analyze() says about dead_tuples in the event of its
estimate increasing our dead_tuples estimate, while being skeptical
(to a varying degree) when it's the other way around.
But now I need to go think about what Andres just brought up...
--
Peter Geoghegan
