ODEProblem(....) vs NeuralODE(....) for neural ODEs

ChrisRackauckas · April 9, 2024, 8:08am

It’s just out of place and ZygoteVJP:

SciML/DiffEqFlux.jl/blob/master/src/neural_de.jl#L49-L55


      
          dudt(u, p, t) = model(u, p)
          ff = ODEFunction{false}(dudt; tgrad = basic_tgrad)
          prob = ODEProblem{false}(ff, x, n.tspan, p)
          
          return (
              solve(prob, n.args...;
                  sensealg = InterpolatingAdjoint(; autojacvec = ZygoteVJP()), n.kwargs...),

Did you try and out of place definition?

function rhs!(u, p, t)
  dudt2(u, p, st)[1]
end

Topic		Replies	Views
Parameter tuning in neural ODE Machine Learning diffeq , flux	15	2077	July 21, 2019
DiffEqFlux with time as additional input to Neural ODE Machine Learning first-steps	7	2268	July 21, 2019
How can I speed up my Neural ODE? Performance question , diffeq , flux , neural-network	4	891	July 20, 2021
Neural ODE fitting really slow Machine Learning diffeq , flux , sciml	13	2029	September 11, 2020
DiffEqFlux: neural_ode stops prematurely General Usage diffeq	4	583	January 24, 2019

ODEProblem(....) vs NeuralODE(....) for neural ODEs

Related topics