I haven't found a single rust crate that can manipulate sets of closed intervals with:
:|
(literally a pick 2 situation)
=> More informations about this toot | View the thread
codata folks: what's a cool name for the opposite of cons (e.g., Container -> Option<T, Container>)?
=> More informations about this toot | View the thread
Key finite domain backtracking search here https://github.com/pkhuong/coppice/blob/602eafd2016a7b8669968c7e5e036049047634f1/src/reverser.rs#L190-L217
=> More informations about this toot | View the thread
I think I finally have a good example https://docs.rs/coppice/latest/coppice/#:~:text=interesting%20about%20Coppice%3A-,the,values%20we%20pass,-. of why I think it's interesting to convert black box functions to branching programs (bit tries).
=> More informations about this toot | View the thread
Sometimes 1.0 just means realising you're done https://crates.io/crates/umash-sys (:
=> More informations about this toot | View the thread
What happened to the data parallel approach to flattening trees into segmented arrays? I'm happy to see renewed interest in depth-first flattening, but I'm wondering if there's a clear motivation for what feels to me like a shift in focus.
=> More informations about this toot | View the thread
what are good (gamedev?) tricks to remember to handle dirty state when you don't want to eagerly trigger callbacks in setters? I'm looking for batching, and ideally no mandatory late binding.
=> More informations about this toot | View the thread
So, the MOAR LAYERS folks like log-sum-exp… do we know if one tried to replace log with a bitcast of float to int, and exp with the reverse? In a float world, the cost is int -> float after bitcast, and float -> int before bitcast.
=> More informations about this toot | View the thread
I felt kind of guilty for hacking something together with bash awk and javascript… but I came back to it 7 years later and I just updated the raw data files, ran one shell script, and it all worked.
=> More informations about this toot | View the thread
Let's say you know the input is an exact multiple of your integer divisor. Does that let you come up with a smaller div-by-mul constant?
https://discuss.systems/@pkhuong/113663255154088174
=> More informations about this toot | View the thread
Does anyone have a container image + build script to run alive-tv (doesn't need the compiler explorer frontend)?
@regehr
=> More informations about this toot | View the thread
Passing around phantom typed proof witnesses as a revolutionary system design idiom that will change the world is peak googler.
=> More informations about this toot | View the thread
Now a pytorch engineer, I guess.
=> More informations about this toot | View the thread
Acar's spin on self-adjusting computations for sliding windows is one of these papers I keep coming back to. 10/10, have read again
Slider: Incremental Sliding Window Analytics.
ACM/IFIP/Usenix Middleware. 2014.
Pramod Bhatotia, Umut A. Acar, Flavio P. Junqueira, Rodrigo Rodrigues.
https://drive.google.com/file/d/1HqOyYBZvVlValpv4PPvoPv-TWzDtDSxD/view
(10 years already!?)
=> More informations about this toot | View the thread
"marks the writer as done" but also CONSUMES THE WRITER BY VALUE!?
https://arrow.apache.org/rust/arrow_array/trait.RecordBatchWriter.html#tymethod.close
=> More informations about this toot | View the thread
My summary: the authors are interested in problems that are naturally seen as finding a set of paths in a graph. The core constraint they're trying to relax is that all the paths must be valid for a large graph. There's also a bunch of side constraints on the paths, with a monotonic structure (sum of weighted edge / vertex <= limit).
They start with a condensed representation of the graph and find lower bound for the lhs coefficients, then iteratively expand the graph to explain what's bad about the most recent solution.
And then you say maybe a Lagrangian relaxation for the side constraints lets you solve with DP instead of bit blasting the whole thing to MIP. Not sure how to scale that to multiple graph constraints (Lagrangian decomposition).
This is intuitively a good fit for multicommodity flow problems… I think some codegen problems where CSE is the dominant factor would also be a good fit? I had an integrated ISel/regalloc based on tree decomposition a long time ago; maybe that sort of thing would also be good candidates, if the condensed graph can have a narrower treewidth.
=> More informations about this toot | View the thread
For the PL <-> OR angle: this could mean that we can find optimal solutions without fully enumerating e-graphs, and instead iteratively feed inequalities to the IP optimiser.
=> More informations about this toot | View the thread
Good haul for this month's optimization online preprints. I have a few things to read, but "Column Elimination: An Iterative Approach to Solving Integer Programs" looks like a nice outer approximation approach that seems intuitive but I never managed to get right. Can't wait to learn how they did it.
https://optimization-online.org/2024/10/column-elimination-an-iterative-approach-to-solving-integer-programs/
=> More informations about this toot | View the thread
gcc silently allowing "0" in asm clobbers seems like such a weird footgun to keep.
=> More informations about this toot | View the thread
Waldorf salad has ~always tasted like nail polish to me. Could it just be that everyone uses stale walnuts?
=> More informations about this toot | View the thread
=> This profile with reblog | Go to pkhuong@discuss.systems account This content has been proxied by September (3851b).Proxy Information
text/gemini