NeuralPDE.jl

NeuralPDE.jl copied to clipboard

[WIP] Generate set of points for each variable

2

https://github.com/SciML/NeuralPDE.jl/issues/469

KirillZubov

[WIP] ForwardDiff AD

22

KirillZubov

I’m trying to solve the dynamics of a massive particle on gravitational force by PINNs. The below code is inspired by an example in the documentation of NeuralPDE package. The...

vnikoofard

Unable to handle non-linear convection terms

4

I am trying to solve the equations for a fully developed laminar flow in a channel (continuity and x-momentum NSE equation) but my PINN code is unable to drive down...

v-chau

Alternative to ModelingToolkit for PDE expressions

10

Hi, Is there an alternative way to describe the PDE to solve that doesn't require the use of ModelingToolkit or Symbolics? For example, if I were to need to write:...

vaishnavtv

Interactive plot script analysis of results

2

Interactive visualization of results would be a convenient tool for analysis Makie.jl looks suitable for this https://github.com/JuliaPlots/Makie.jl examples: http://juliaplots.org/MakieReferenceImages/gallery/index.html  two simple cases: iteration number slider ->(prediction and error in...

KirillZubov

Correct sampling

2

Closes #609

YichengDWu

QuasiRandomTraining(sampling_alg = SobolSample()) is not working correctly

18

```julia julia> s = NeuralPDE.generate_quasi_random_points(300, [[-1.0,-2.0],[1.0,2.0]], Float64, LatinHypercubeSample()) 2×300 Matrix{Float64}: -1.08 -1.33333 -1.86 -1.49333 … -1.54667 -1.66667 -1.47333 1.35667 1.53667 1.84333 1.07333 1.28 1.66667 1.24667 julia> scatter(s[1,:],s[2,:]) ``` ```julia julia>...

YichengDWu

NN ODE loss gets squared

1

I think that the objective functions/loss functions used for `GridTraining` and `StochasticTraining` generated in [ode_solve.jl](https://github.com/SciML/NeuralPDE.jl/blob/master/src/ode_solve.jl) "get squared" before they are passed to `OptimizationFunction`, meaning, the function $(\sum_i \ell_i^2)^2$ is passed...

milzj

Treating states properly, second encounter

3

I just realized that the recent PR #604 did not fix anything. As `Phi` is a mutable field of immutable `PhysicsInformedNN`. The related source code is a bit messy. Adapting...

YichengDWu