[slack] <Arno Strouwen> The surrogate optimizer page is missing a few examples with all the different
http://surrogates.sciml.ai/dev/optimizations/
Also the formatting is not very uniform over the different methods, similarly so for /dev/surrogate/ and /dev/sample/ .
NewOptimizatonType names in them. I can guess that Expected Improvement is going to be EI(), but it could be nice to add them.http://surrogates.sciml.ai/dev/optimizations/
Also the formatting is not very uniform over the different methods, similarly so for /dev/surrogate/ and /dev/sample/ .
[slack] <ludoro> Would you care to open an issue with your suggestions?
[slack] <Arno Strouwen> I will do that. But I also have some other questions:
[slack] <Arno Strouwen> If I remember correctly there used to be an example in the docs linking another kriging library, where the surrogate was fitted using maximum likelihood. I can no longer find this example, I found the ML estimation much more robust than tweaking the kriging hyperparameters in Surrogates.jl. The main page still references this functionality.
[slack] <ludoro> Oh that’s strange, i think you are referring to Stheno.jl though
[slack] <Arno Strouwen> Yes I think it was Stheno. There is still one reference to that in the docs, but nothing more.
[slack] <ludoro> On the Hyperparameter optimization, I plan on having a tutorial on that showing how to use surrogates with some common libraries.
[slack] <ludoro> Thanks for your suggestions!
[slack] <Arno Strouwen> One more question: Any suggestions which surrogate and Bayesian optimizer is best when you have a noisy objective, because it involves (Markov-chain) Monte-Carlo integration?
[slack] <chrisrackauckas> In about 45 minutes (8:30pm EST) I'm doing a quick seminar talk which will discuss some of the stuff with universal differential equations and adjoints of differential equations. If anyone wants to join you can find a zoom link here: https://www.cmu.edu/aced/sciML.html
[slack] <briochemc> What's the correct way to check if a function is inplace? I.e., in
```julia> foo(du, u, p, t) = "in place"
foo (generic function with 1 method)
julia> ODEFunction(foo)
(::ODEFunction{true,typeof(foo),LinearAlgebra.UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}) (generic function with 7 methods)``
how does ODEFunction assigniip = true`? Not sure where to look for the code that does that 😅
[slack] <briochemc> I'm guessing it checks
nargs, but I want to be sure how it's supposed to be done 🙂
[slack] <briochemc> (also a
nargs search on DiffEqBase gave me no results 🤔)
[slack] <chrisrackauckas> it should?
[slack] <dalarev> An explicit output, you mean?: https://files.slack.com/files-pri/T68168MUP-F01FDSPFD52/download/image.png
[slack] <mschauer> I looked at idrs and it would require some thinking to tell if this is legal for overdetermined systems for example
[slack] <dalarev>
A\b works, but I am benchmarking different methods.
[slack] <mschauer>
A\b does something perhaps overly smart and switches the solver depending on the dimensions i believe
[slack] <dalarev> https://discourse.julialang.org/t/efficient-way-of-doing-linear-regression/31232/33 post tells me it's probably the most accurate, but my
A matrix comprises derivative operators, so I'm also looking at DiffEqOperators.
[slack] <mschauer> and then perhaps with a solver which uses the Hermitian structure… ah, you just said so much
[slack] <chrisrackauckas> you can rerun https://benchmarks.sciml.ai/html/MultiLanguage/wrapper_packages.html
[slack] <mschauer> Well, can you do QR? Then do it. I assumed that you use an iterative solver because you must
[slack] <David Millard> Hi, I have a second order ODE in a classical mechanics context. Is
SecondOrderODEProblem still supported? I had a hard time understand what inputs the dynamics function should take. Unfortunately reading https://github.com/SciML/DiffEqBase.jl/blob/864c379ead274243acdd965bd8942d1da1deb3bb/test/problem_creation_tests.jl#L51 didn't shed a lot of light for me. I'm not sure what the relationship between u and v are, and the docstring args from the SecondOrderODEProblem constructor seems flipped compared to its implementation. Thanks for any information!
[slack] <chrisrackauckas> @David Millard https://tutorials.sciml.ai/html/models/01-classical_physics.html should help
[slack] <chrisrackauckas> RecursiveArrayTools.ArrayPartition
[slack] <jonas.isensee> Thanks you, that seems to work.
Not sure that is much better for my particular usecase though.
I want to work with the matrix directly andu.x[2] is probably not much better than`@view u[:,2:4]`
at the cost of introducing another abstraction layer.
(relevant since this is for teaching...)
[slack] <chrisrackauckas> LabelledArrays.jl is another.
[slack] <jonas.isensee> Thanks for your help
[slack] <chrisrackauckas> no problem