Vincent Lostanlen
Vincent Lostanlen
@MuawizChaudhary LMK if/when you need a new review
@MuawizChaudhary yes, good. nice work
this is nice but should be merged after v0.3 let's talk about it tomorrow
yo @cyrusvahidi @MuawizChaudhary now that we have a spin PR #922 i'd like to revive this one mind if i take this code and make a fresh PR? (with credit)
By "batched" do you mean breadth-first search, as opposed to depth-first search? We agreed on depth-first when designing Kymatio at the start for reasons of memory management. Indeed, a serious...
@MuawizChaudhary IIRC the ratio between BFS and DFS is at least `Q1*Q2` in 1D, and at least `L1*L2` in 2D. This can be significant in audio, in which one routinely...
> This way, if power users want a faster implementation, they can code their own breadth first search core, and then replace ScatteringObject.scattering#d with the new implementation. Right. I think...
Hello @buffy-nevc, This is a good feature request indeed. I have been thinking about the issue (in 1-D) and it seems relatively straightforward, although i haven't tried it. With Morlet...
@edouardoyallon it's less a problem of native/non-native complex numbers than about making sure that we cover the "negative frequency range" [pi, 2*pi[ in the first layer. So we definitely need...
Thank you for opening this issue. 1. If we are to have stem-level `has_bleed` annotations, then I cannot find a use case for keeping it at the track level as...