Skip to main content

Browse the archive

https://github.com/JuliaDiffEq/DiffEqFlux.jl
13 April 2020, 12:04:22 UTC
Revision 9af1cd7313fe1b3e7d235bc9f42212961ae93a1e authored by Christopher Rackauckas on 08 April 2020, 18:36:50 UTC, committed by GitHub on 08 April 2020, 18:36:50 UTC 
Merge pull request #220 from ranjanan/RA/quaddirect
 
Add QuadDIRECT global optimizer
2 parent s 35cfab2 + 424703e
Raw File
Tip revision: 9af1cd7313fe1b3e7d235bc9f42212961ae93a1e authored by Christopher Rackauckas on 08 April 2020, 18:36:50 UTC
Merge pull request #220 from ranjanan/RA/quaddirect
Tip revision: 9af1cd7
newton_neural_ode.jl 
using DiffEqFlux, Flux, Optim, OrdinaryDiffEq

n = 5  # number of ODEs
tspan = (0.0, 1.0)

d = 5  # number of data pairs
x_data = rand(n, 5)
y_data = rand(n, 5)
training_data = Iterators.cycle([(x_data[:, i], y_data[:, i]) for i in 1:d])

NN = Chain(Dense(n, 10n, tanh),
           Dense(10n, n))

nODE = NeuralODE(NN, tspan, ROCK4(), reltol=1e-4, saveat=[tspan[end]])

loss_function(θ, x, y) = Flux.mse(y, nODE(x, θ))
res = DiffEqFlux.sciml_train(loss_function, nODE.p, LBFGS(), training_data)
res = DiffEqFlux.sciml_train(loss_function, nODE.p, NewtonTrustRegion(), training_data)

NN = FastChain(FastDense(n, 10n, tanh),
               FastDense(10n, n))

nODE = NeuralODE(NN, tspan, ROCK4(), reltol=1e-4, saveat=[tspan[end]])

loss_function(θ, x, y) = Flux.mse(y, nODE(x, θ))
res = DiffEqFlux.sciml_train(loss_function, nODE.p, LBFGS(), training_data)
res = DiffEqFlux.sciml_train(loss_function, nODE.p, NewtonTrustRegion(), training_data)
back to top