[keysersosa]

Confirmed, but the time dependent model didn't ask for long. I seem to recall switching to the current model (with slightly different constants) during the switch to python in late '05.

That said, one of the early "mistakes" we made was not noticing the obvious step of taking the log of on the exponential form formula, which ended up meaning we had only a couple hundred days before we started to run afoul of the max float size in postgresql. I also seem to recall coming up with that obvious trick with only a few days to spare.

[jedberg]

I'm really glad you're here, because I was about to comment that you're the only person I know who understands this stuff.

(for those that don't know keysersosa was the guy who wrote the current hot algorithm)

[tim333]

Ah - I was thinking about how exponential growth would go over the max size and how to deal with it and also wondering why they were taking logs and not putting two and two together. It's surprising how tricky it can be to see what seems obvious with hindsight.