Closed amckay1 closed 4 years ago
The plot of the data looks transposed.
As for not converging, we just updated the docs to 1.0 and this may have a bug. My guess is that I accidentally didn't pass the parameters somewhere
FWIW we know it works because NeuralNetDiffEq works, and it uses the SDE based method for solving high dimensional PDEs, so it's some silly doc issue
Definitely, not high priority, just wanted to let you know as the new paper drives more traffic to the docs. And I agree it looks transposed, but probably because the x-axis is not responding correctly to the timespan specified by ensemble_nsum
, and I think if you zoom in on the 0.0:1.0 range you'll see that the sde_data
is plotting correctly. Maybe something to do with the plots
recipe for ensemble_nsum
? Speculating, will keep exploring...
Fixed! It had a plot recipe issue in DiffEqBase mixed with a small thing I forgot to handle in the Zygote transition (the test had an @test_broken
on it to remind me... oops). All is good on master now and I'll make a patch release soon.
Wonderful, thank you!
Very excited to try out some of the demo code for UDEs, and I didn't have any trouble with the ODE example (starting at https://github.com/JuliaDiffEq/DiffEqFlux.jl#universal-differential-equations) but when I tried to use the SDE example (starting at https://github.com/JuliaDiffEq/DiffEqFlux.jl#neural-sde-example) I encountered a couple issues:
I'm guessing 1 and 2 are linked, and possibly a versioning issue: I'm updated on everything but not on master with DiffEqFlux v1.0.0, StochasticDiffEq v6.17.1, Plots v0.28.4, Flux v0.10.1, and DiffEqBase v6.12.2.
Will try to debug it more later if I can :)
Exact code ran in fresh Julia v1.3.1 session (OSX) below (copied from README.md):