Quick Start

Install the library

pip install neural-diffeqs

Import the library

import neural_diffeqs

Data

Let’s say you have some temporally-resolved data, spanning three time points. There are 200 12-dimension samples at each time point.

# the initial state
X0 = torch.randn([200, 12])

# later states
X1 = torch.randn([200, 12])
X2 = torch.randn([200, 12])

time = torch.Tensor([1, 4, 7])

Neural network function

Call together an SDE (or ODE):

SDE = neural_diffeqs.NeuralSDE(
    state_size = 12, mu_hidden = [32, 32], sigma_hidden = [32, 32],
)

ODE = neural_diffeqs.NeuralSDE(state_size = 12, mu_hidden = [32, 32])

Make a prediction

import torchsde

# dt is an important parameter to tune; we'll start with 0.1

X_hat_sde = torchsde.sdeint(SDE, X0, ts = time, dt = 0.1)

X_hat_ode = torchsde.sdeint(ODE, X0, ts = time, dt = 0.1)

# compare to X1, X2 (observed states)
# integrate into a training regimen to fit the mu, sigma networks