Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • 15:38
    ChrisRackauckas commented #309
  • 15:38
    ChrisRackauckas reopened #309
  • 15:38
    ChrisRackauckas closed #309
  • 15:37

    julia-tagbot[bot] on v6.9.1

    (compare)

  • 15:32
    JuliaRegistrator commented #294
  • 15:32
    ChrisRackauckas commented #294
  • 15:32

    ChrisRackauckas on master

    Update Project.toml (compare)

  • 15:31

    ChrisRackauckas on docstrings

    (compare)

  • 15:31

    ChrisRackauckas on master

    adds docstrings for callbacks a… minor link fixes Merge pull request #399 from Ju… (compare)

  • 15:31
    ChrisRackauckas closed #399
  • 14:36
    kanav99 commented #309
  • 14:36
    kanav99 commented #309
  • 14:35
    kanav99 synchronize #309
  • 14:34
    kanav99 ready_for_review #309
  • 14:31
    kanav99 synchronize #399
  • 14:31

    kanav99 on docstrings

    minor link fixes (compare)

  • 14:29
    kanav99 synchronize #309
  • 13:43
    Vaibhavdixit02 commented #283
  • 13:42
    Vaibhavdixit02 commented #283
  • 13:33
    ChrisRackauckas closed #283
BridgingBot
@GitterIRCbot
[slack] <chrisrackauckas> Maybe I just let it run for a week and see what comes out.
[slack] <amin.sadeghi> I have not yet implemented it using DiffEq, but eventually yes it's going to be one
[slack] <isaacsas> I’ll try to locally time just Kencarp4 to see what I get for the different options. Then we’d at least have a ballpark timing estimate.
[slack] <chrisrackauckas> oh, you're implementing JuliaDiffEq/SteadyStateDiffEq.jl#15 ?
[slack] <amin.sadeghi> Didn't know about that but it seems yes!
[slack] <chrisrackauckas> yeah I'd look into the literature and see how they handle this relaxation
[slack] <amin.sadeghi> I found this so far, but I think it might be too specific to the PDE they're solving: https://link.springer.com/content/pdf/10.1007/s00466-008-0255-5.pdf
BridgingBot
@GitterIRCbot
[slack] <amin.sadeghi> but maybe after all, it is highly dependent on the PDE?
[slack] <chrisrackauckas> it shouldn't be. The two papers in the repo are generic methods IIRC
[slack] <amin.sadeghi> I'll read up those papers then
[slack] <amin.sadeghi> Thanks
BridgingBot
@GitterIRCbot

[slack] <rveltz> Hi,

Is there a way to pass parameter to a callback function? I guess I can pass it to integrator but say I want to modify somebody else prob to add a callback and that the callback parameter has nothing to do with the one in prob. How can I achieve this?

[slack] <chrisrackauckas> usually it would just be `integrator.p``
[slack] <chrisrackauckas> you can make it a struct to keep different parameters separate?
BridgingBot
@GitterIRCbot
[slack] <rveltz> yes I guess and use remake to alter differenly the parameters of the vector field and callback condition?
[slack] <chrisrackauckas> yeah, that or modify p
[slack] <chrisrackauckas> I think we need a better parameter interface
[slack] <chrisrackauckas> but that's a whole story.
[slack] <rveltz> modify p? What do you mean?
BridgingBot
@GitterIRCbot
[slack] <isaacsas> OK, latest BCR kencarp4 timings on my computer (no autodiff yet):
 Section               ncalls     time   %tot     avg     alloc   %tot      avg
 ──────────────────────────────────────────────────────────────────────────────
 DenseJac Eval1             1     487s  17.7%    487s   7.19GiB  7.42%  7.19GiB
 SparseJac Eval1            1     470s  17.1%    470s   6.57GiB  6.78%  6.57GiB
 KenCarp4-FD-2              1     439s  15.9%    439s   0.99GiB  1.02%  0.99GiB
 KenCarp4-FD-1              1     425s  15.4%    425s   3.10GiB  3.20%  3.10GiB
 KenCarp4-Sparse-1          1     258s  9.35%    258s   35.4GiB  36.5%  35.4GiB
 KenCarp4-Sparse-2          1     255s  9.24%    255s   35.0GiB  36.1%  35.0GiB
 KenCarp4-Analytic-1        1     217s  7.87%    217s   4.52GiB  4.66%  4.52GiB
 KenCarp4-Analytic-2        1     164s  5.96%    164s    943MiB  0.95%   943MiB
 ODERHS Eval1               1    41.0s  1.49%   41.0s   3.23GiB  3.34%  3.23GiB
 DenseJac Eval2             1   1.07ms  0.00%  1.07ms     32.0B  0.00%    32.0B
 ODERHS Eval2               1    153μs  0.00%   153μs     32.0B  0.00%    32.0B
 SparseJac Eval2            1    123μs  0.00%   123μs     32.0B  0.00%    32.0B
 ──────────────────────────────────────────────────────────────────────────────
[slack] <isaacsas> FD = finite difference. Spare and Analytic are using the explicit routines to fill in the matrices from DEBio.
[slack] <isaacsas> So I guess the issue on the server must be autodiff since these don't seem too bad. (This is using tstop=100000 as in the benchmark.)
[slack] <isaacsas> and abs/rel tols of 1e-8
BridgingBot
@GitterIRCbot
[slack] <yingbo_ma> How long does Sundials take?
[slack] <yingbo_ma> Do you mind to also try TRBDF2 ?
BridgingBot
@GitterIRCbot
[slack] <wsphillips> Likely simple answer, but I haven't seen a clear example in the tutorials:
How are state dependent parameters defined when working with a ParameterizedFunction @ode_def? In this case, the gating variables on a Hodgkin-Huxley system are dependent on the current value of V.
[slack] <chrisrackauckas> ParameterizedFunctions cannot handle this
[slack] <chrisrackauckas> it's somewhat of a deprecated library though
[slack] <wsphillips> ah
[slack] <chrisrackauckas> it works for what it does, but it's limited
[slack] <chrisrackauckas> ModelingToolkit.jl is really where all of the work has gone
[slack] <chrisrackauckas> and it can support htis.
[slack] <wsphillips> ok, then that's where I will head off too 😃
BridgingBot
@GitterIRCbot
[slack] <isaacsas> I'll give it a try later tonight once I'm back at my computer.
[slack] <isaacsas> Along with Sundials.
[slack] <yingbo_ma> Thank you!
BridgingBot
@GitterIRCbot
[slack] <wsphillips> Can Callbacks and the CallbackLibrary be mixed into ModelingToolkit systems?
BridgingBot
@GitterIRCbot
[slack] <chrisrackauckas> You can use them on the functions it generates
BridgingBot
@GitterIRCbot
[slack] <wsphillips> Got it! I am officially in business 🤓 Thanks for the pointers--may not seem like much but probably saved me quite a bit of struggle.
BridgingBot
@GitterIRCbot
[slack] <chrisrackauckas> no proble mat all
BridgingBot
@GitterIRCbot
[slack] <wsphillips> So, if I understand correctly, with the generated function from ModelingToolkit.jl, I can not only use it to build an ODEProblem -> solve using DiffEq solvers, but that the idea is I could pass that same problem into DiffEqFlux to solve its parameters standalone or in combination with any neural net layers?
[slack] <chrisrackauckas> yeah, we will be improving that connection soon as well.
[slack] <wsphillips> yeah, I imagine that's not trivial...but awesome stuff
[slack] <wsphillips> definitely a change from when I learned the basics in grad school using XPPAUT
BridgingBot
@GitterIRCbot
[slack] <rveltz> you mean on my side ?
[slack] <rveltz> if this works, we can have stiff ode on the GPU?
[slack] <rveltz> As for preconditioners, the left one is SUPER easy to do. I will bump jutho
BridgingBot
@GitterIRCbot
[slack] <rveltz> we basically need a proof of concept
BridgingBot
@GitterIRCbot
[slack] <chrisrackauckas> ForwardDiff Jacobians work with the GPU
[slack] <chrisrackauckas> so this isn't necessary, though could be better in some circumstances.