Anthony Blaom, PhD
Anthony Blaom, PhD
This also confused me for a bit.
Other issues: - all packages have to be loadable at the same time; so if one is behind in a compat update, this can preclude updating some other package's metadata...
Has there been any recent progress on this issue?
@darsnack Thanks for the update and speed response. > Is (2) required if you know a priori that every model in Metalhead has these switches? No.
I see 0.8 has been released. But I take it this is still open?
Nice observation. May be tricky to fix quickly. I'm curious how you came to identify the issue. Do you have a concrete use case with performance substantially compromised by this...
Thanks @MilesCranmer for the further details and the very nice idea for a fix. Perhaps after I have a chance to look at a direct fix we can go this...
Okay, the type instability in `last_model` is an easy fix. I just need to annotate the `old_model` field in the `Machine` struct. But the benchmarks are still bad. Tried `Chulhu.jl`...
I have made some more progress removing some type instability with predict. Will cross-reference to a PR shorty. However, while this reduces bloat in `predict(::Machine, ...)` across the board, it...
@juliohm Friendly reminder.