You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hey @ricardoV94, for this issue I followed the PR for kv and I expressed iv in the form of a ive helper as iv(v,z) = ive(v,z) * exp(z), but it overflows for iv(v,z) when z>100 for v=4.5 that is around 1e36 which I don't think should be happening.
Please let me know where I'm going wrong. Thanks!
Sorry for missing this thread. How long after does iv itself start overflowing? It's not completely surprising that the multiplication will be a bit less precise.
It might or not be a deal breaker.
The rewrite for the log will probably still make things more stable via the ive route
For now I think it's fine trade-off for simpler codebase. We can revisit later. Most applications will probably want to work on log-scale anyway, where the ive approach should win clearly
Activity
AdvH039 commentedon Dec 7, 2024
Hey @ricardoV94, for this issue I followed the PR for kv and I expressed iv in the form of a ive helper as iv(v,z) = ive(v,z) * exp(z), but it overflows for iv(v,z) when z>100 for v=4.5 that is around 1e36 which I don't think should be happening.
Please let me know where I'm going wrong. Thanks!
ricardoV94 commentedon Jan 23, 2025
Sorry for missing this thread. How long after does iv itself start overflowing? It's not completely surprising that the multiplication will be a bit less precise.
It might or not be a deal breaker.
The rewrite for the log will probably still make things more stable via the ive route
ricardoV94 commentedon Jan 31, 2025
I'm not seeing a very early overflow?
It overflows a tiny bit before iv:
For now I think it's fine trade-off for simpler codebase. We can revisit later. Most applications will probably want to work on log-scale anyway, where the ive approach should win clearly