ChrisRackauckas

33 ms·

301.
by ChrisRackauckas 5y ago
> Despite the comments above about Julia having this already, Awkward Array is a rather different thing from JIT-compiled data structures, and it would be as valuable in Julia as it is in Python, least of all because GPUs have to be add
302.
by ChrisRackauckas 5y ago
Have you used almost any ODE solver? Almost every single one uses embedded methods to adapt time steps. ode23, ode45, ode23t, ode23tb, ode15s, LSODE, LSODA, radau, rodas, VODE, CVODE, ... even for DAEs you have DASSL, IDA, ... I can keep go
303.
by ChrisRackauckas 5y ago
That's precisely what composability solves. We're seeing in CLIMA that using more general highly optimized solvers can greatly decrease the `f` cost count moreso than focusing on really low level optimizations. Especially in thing
304.
by ChrisRackauckas 5y ago
Yes, that has nothing to do with what I just said though. Of course MD is fine because symplectic ODE solvers cannot generally have adaptivity (without tricky and very expensive handling of `t` inside of the Hamiltonian which nobody does be
305.
by ChrisRackauckas 5y ago
XLA only optimizes quasi-static code, which does not include adaptive numerical solvers like those for ODEs. It's a generally good assumption for ML though, but there are ways to break it. I wrote a piece showcasing some ideas around t
306.
by ChrisRackauckas 5y ago
No, in fact, you want to go lower order with lower precision. The real answer is that if the solution is in the chaotic regime then maybe Float16 is fine because you'll be dominated by other numerical errors anyways (if you're als
307.
by ChrisRackauckas 5y ago
Oh another thing, in the chaotic regime which is of interest, standard automatic differentiation schemes don't even apply as you require shadow adjoints given the shadow trajectory leads to inaccurate calculations for the gradient. Jul
308.
by ChrisRackauckas 5y ago
The real problem with the Jax code is that the non-composable programming language setup put it into a corner where it's using an extremely inefficient time stepping method that it has "optimized", but how is it optimized if
309.
by ChrisRackauckas 5y ago
That's an old example. It will now default to Enzyme and should do quite a bit faster. I should update that.
310.
by ChrisRackauckas 5y ago
Almost all of the improvements came from changes to Julia's Base that improved compilation times and changes to tooling which suggest trivial changes to improve compilation times. In v1.5 or so this would've been a heroic feat. In
311.
by ChrisRackauckas 5y ago
> namely that gradients are not all that helpful when your optimization landscape is insanely non-linear Did you try multiple shooting? https://diffeqflux.sciml.ai/dev/examples/multiple_shooting/ I want to
312.
by ChrisRackauckas 5y ago
>I hope those benchmarks are coming in hot M1 is extremely good for PDEs because of its large cache lines. https://github.com/SciML/DiffEqOperators.jl/issues/407#issue... The JuliaSIMD tools which are inte
313.
by ChrisRackauckas 5y ago
Let's even put raw numbers to it. DifferentialEquations.jl usage has seen compile times drop from 22 seconds to 3 seconds over the last few months. https://github.com/SciML/DifferentialEquations.jl/issues&#x2F
314.
by ChrisRackauckas 5y ago
>One of the domains where I hope Julia will excel is precisely as you comment on physics + ml (I check frequently if MuJoCo source code is available at last). Somebody already did a comparison of DifferentialEquations.jl's standard
315.
by ChrisRackauckas 5y ago
It's acknowledged. The full redefinition of any function in any context was solved by RuntimeGeneratedFunctions and we use it extensively throughout SciML, ModelingToolkit, Pumas, etc. so it's fair to say struct redefinitions are
316.
by ChrisRackauckas 5y ago
Okay, so then use RuntimeGeneratedFunctions+Revise. Done. Julia can do it, but having a setup so that people can do instant review of changes on the live production system is still a little bizarre and I wouldn't do it even though we c
317.
by ChrisRackauckas 5y ago
We legally cannot do things like say live code change the production system underlying the analysis of ongoing clinical trials without going through the whole software development lifecycle required by the FDA (I.e. the Pumas example again)
318.
by ChrisRackauckas 5y ago
Yup exactly. I hate these super general statements like "better ecosystem". These blanket statements obviously make no sense since you can pull out 20,000 counter examples off the top of your head. Instead of "Go beats Julia
319.
by ChrisRackauckas 5y ago
yupyupyup, you got it. :)
320.
by ChrisRackauckas 5y ago
>Go beats Julia in parts where python is not good at. I have not seen good results from differential equation solvers in Go.
321.
by ChrisRackauckas 5y ago
Call me a cynic but I would hope almost no devs had the power to live update the production codebase. That sounds crazy to me. To me, that kind of thing should only be done with (approved/verified) versions which are cut and tested bef
322.
by ChrisRackauckas 5y ago
> Zero based indexing makes zero sense The sensibility of the index choice is equal to the starting value of the index.
323.
by ChrisRackauckas 5y ago
The person who comes to a Julia discussion and says "so, how's 1-based indexing?" is the same person who walks up to people in the office and goes "so, how's the weather today?" every single day. It's the
324.
by ChrisRackauckas 5y ago
You're not doing interactive development on production. Production is the case where you just build a system image and deploy it without compile times. This is done all of the time, for example with Pumas.
325.
by ChrisRackauckas 5y ago
> `Σλ∀φ` is a chore to read for me. These letters only make sense if I've absorbed the author's notation beforehand, like in a math proof. It's a hell of a lot better than reading `Sigma`, `lambda`, etc. and having to do b
326.
by ChrisRackauckas 5y ago
Not in real-world contexts. This is spelled out in Julia for Biologists ( https://arxiv.org/abs/2109.09973 ) which does the operation counting to show why using Numba with SciPy is still an order of magnitude slower in s
327.
Using machine learning to derive black hole motion from gravitational waves(phys.org)
2 points by ChrisRackauckas 5y ago|0 comments
328.
by ChrisRackauckas 5y ago
There are trade-offs made. This discussion of "why Julia" describes how multiple dispatch + type stability is what gives the speed, but the trade-offs and edge cases associated with that. http://ucidatascienceinitiative
329.
by ChrisRackauckas 5y ago
Unitful.jl
330.
by ChrisRackauckas 5y ago
For more information on the backstory behind this method of analyses, see the Twitter thread https://twitter.com/ChrisRackauckas/status/14541131552783237... . Importantly, I want to emphasize that this is about a q

More ›