cbkeller
Joined 8 years ago
Karma
1,804
Submissions
407
Member since
Oct 23
geologist - brenhinkeller.github.io
View on Hacker NewsRecent submissions
Comment · 2 years ago
It looks like you very dramatically missed the point
Comment · 3 years ago
So in other Julia geometry-related projects that may be true, but for this particular corner of the ecosystem the main author (Júlio Hoffimann) has actually implemented much of the underlying geometry and other code from scratch (to the best of my understanding) in pure Julia in a whole set of packages, including e.g.
https://github.com/JuliaGeometry/Meshes.jl https://github.com/JuliaGeometry/Rotations.jl https://github.com/JuliaEarth/GeoStatsBase.jl https://github.com/JuliaEarth/PointPatterns.jl
Comment · 3 years ago
For this crowd, "machine learning" should really be read as parallel Bayesian inversion via MCMC (Metropolis modified with a form of "parallel tempering"), but that was a bit much to explain to a popular audience.
The underlying paper is https://doi.org/10.1126/science.adh3875: Cox, Alexander A., and C. Brenhin Keller. "A Bayesian inversion for emissions and export productivity across the end-Cretaceous boundary." Science 381.6665 (2023): 1446-1451.
That paper is more focused on the general concepts than the implementation, but FWIW the new code is in Julia and in the supp mat, and calls an existing C program called LOSCAR for the forward modelling.
Comment · 3 years ago
I suspect there's a bit of an Eternal September effect going on as a wider audience ends up here (possibly fleeing the continuing "enshittification" of most all for-profit online fora)
Comment · 3 years ago
Femptolisp, a "Scheme dialect"
Comment · 3 years ago
That's how it works in any language on a supercomputer. MPI is pretty much the only game in town for inter-node communication.
Comment · 3 years ago
The most IDE-ish experience is probably currently the vscode plugin. I haven’t used pycharm specifically for comparison, but I suspect “probably not” as compared to a polished paid IDE in general. There’s progress being made though.
Comment · 3 years ago
> While the details of CCS provisioning vary in the different versions of the GPL agreements, the general principle is that CCS need to be provided either (a) along with the binary distributions to those who receive, or (b) to those who request pursuant to a written offer for source.
Since the written offer is apparently mandatory [1], does this mean that a potential way forward (if Red Hat intends to not break the GPL) is for Rocky and Alma to make regular written requests for source?
[1] https://www.gnu.org/licenses/gpl-faq.html, specifically
> If you commercially distribute binaries not accompanied with source code, the GPL says you must provide a written offer to distribute the source code later. When users non-commercially redistribute the binaries they received from you, they must pass along a copy of this written offer. This means that people who did not get the binaries directly from you can still receive copies of the source code, along with the written offer.
> The reason we require the offer to be valid for any third party is so that people who receive the binaries indirectly in that way can order the source code from you.
Comment · 3 years ago
This is pretty clearly against the spirit, albeit probably not the letter, of the GPL. Sad to see.
Comment · 3 years ago
The author links this in the fourth paragraph; their argument is that while compute is not a viable moat, training data may be.
Comment · 3 years ago
Yes, I think so (and many already have)
Comment · 3 years ago
I've seen this criticism that it's "difficult to reason about which method will actually be called because of the type system" once before, but it absolutely flummoxes me... I have literally never been confused about which method is going to be called by my code, and I'm not even a proper computer scientist (just a regular scientist scientist).
Maybe this is just an issue of not having years of OO habits influencing the way I reason about dispatch in Julia?? Honestly not sure.
Comment · 3 years ago
It has good power/weight ratio in theory, esp. if you're burning it (c.f. space shuttle), but hard to store or transport it both densely and safely in a practical way -- i.e., unless you're comfortable dragging around cryogenic lH2 in your sports car
Comment · 3 years ago
Seasonal storage is an interesting idea. Rare grid outages seems pretty easy with batteries if the South Australia example is anything to go by though, and for medium-term storage there's also pumped hydro -- not sure where that compares in cost?