Stochastic Simulations

This vignette demonstrates how to use the simulate method for calculating k-step (state and observation) simulations.

Notation

---

Let the set of observations from the initial time $t_0$ until the current time $t_{i}$ be noted by $$\mathcal{Y}_{i} = \left\{ y_{i}, y_{i-1},...,y_{1},y_{0}\right\}$$

A k-step simulation is a sample of the stochastic path of the model stochastic differential equation k time-steps into the future, conditioned on the current state estimate with mean and covariance $$ \hat{x}_{i|i} = \mathrm{E}\left[ x_{t_{i}} | y_{t_{i}} \right] \\ P_{i|i} = \mathrm{V}\left[ x_{t_{i}} | y_{t_{i}} \right] $$ A single stochastic simulation can be obtained using the Euler-Maruyama scheme by $$X_{t_{j+1}} = X_{t_{j}} + f(X_{t_{j}},u_{t_{j}},t_{j}) \, \Delta t_{j} + G(X_{t_{j}},u_{t_{j}},t_{j}) \, \Delta B_{j}$$ for $j=i,...,i+k$, where the initial point follows $$X_{t_{i}} \sim N(\hat{x}_{i|i}, P_{i|i} )$$ and $$\Delta B_{j} \sim N(0,\Delta t_{j})$$

Arguments

---

The simulate method accepts the following arguments

model$simulate(data,
               pars = NULL,
               use.cpp = FALSE,
               method = "ekf",
               ode.solver = "rk4",
               ode.timestep = diff(data$t),
               simulation.timestep = diff(data$t),
               k.ahead = nrow(data)-1,
               return.k.ahead = 0:k.ahead,
               n.sims = 100,
               initial.state = self$getInitialState(),
               estimate.initial.state = private$estimate.initial,
               silent = FALSE)

Argument: pars

---

See the description in the predict vignette.

Argument: use.cpp

---

See the description in the predict vignette.

Argument: method

---

See the description in the estimate vignette.

Note: The simulate method is currently only available using the Extended Kalman filter (method="ekf).

Argument: ode.solver

---

See the description in the estimate vignette.

Note: When the argument use.cpp=TRUE then the only solvers available are euler and rk4.

Argument: ode.timestep

---

See the description in the estimate vignette.

Argument: k.ahead

---

See the description in the predict vignette.

Argument: return.k.ahead

---

See the description in the predict vignette.

Argument: simulation.timestep

---

This argument is the same as ode.timestep but determines the time-steps used between data-points when performing the Euler-Maruyama simulation.

Argument: n.sims

---

The number of stochastic simulations (trajectories) generated.

Argument: initial.state

---

See the description in the predict vignette.

Argument: estimate.initial.state

---

See the description in the predict vignette.

Argument: silent

---

See the description in the predict vignette.

Example

We consider a modified Ornstein Uhlenbeck process:

$$ \mathrm{d}x_{t} = \theta (a_t - x_{t}) \, \mathrm{d}t \, + \sigma_{x} \, \mathrm{d}b_{t} \\ y_{t_{k}} = x_{t_{k}} + \varepsilon_{t_{k}} $$ where the mean is some complex time-varying input $a_t = tu_{t}^{2}-\cos(tu_{t})$, and $u_{t}$ is a given time-varying input signal.

We create the model and simulate the data as follows:

model = ctsmTMB$new()
model$addSystem(dx ~ theta * (t*u^2-cos(t*u) - x) * dt + sigma_x*dw)
model$addObs(y ~ x)
model$setVariance(y ~ sigma_y^2)
model$addInput(u)
model$setParameter(
  theta   = c(initial = 2, lower = 0,    upper = 100),
  sigma_x = c(initial = 0.2, lower = 1e-5, upper = 5),
  sigma_y = c(initial = 5e-2)
)
model$setInitialState(list(1, 1e-1*diag(1)))

# Set simulation settings
set.seed(20)
true.pars <- c(theta=20, sigma_x=1, sigma_y=5e-2)
dt.sim <- 1e-3
t.sim <- seq(0, 1, by=dt.sim)
u.sim <- cumsum(rnorm(length(t.sim),sd=0.1))
df.sim <- data.frame(t=t.sim, y=NA, u=u.sim)

# Simulate data
sim <- model$simulate(data=df.sim, 
                      pars=true.pars, 
                      n.sims=1,
                      silent=T)

# Grab first simulation trajectory
x <- sim$states$x$i0$x1

# Extract observations from simulation and add noise
iobs <- seq(1,length(t.sim), by=10)
t.obs <- t.sim[iobs]
u.obs <- u.sim[iobs]
y = x[iobs] + true.pars["sigma_y"] * rnorm(length(iobs))

# Create data-frame
.data <- data.frame(
  t = t.obs,
  u = u.obs,
  y = y
)

We can simulate many trajectories using:

sim <- model$simulate(data=.data, 
                      pars=c(20,1,0.05), 
                      n.sims=100,
                      silent=T)

with parameters $\theta = 20, \sigma_{x} = 1, \sigma_{y} = 0.05$.

Note: The value of $\sigma_{y}$ has no impact when doing “full” simulations (i.e. choosing maximum k.ahead) since no data updates occur.

The simulations can be plotted quickly using matplot:

# Get the first (and only in this case) k-step simulation data.frame
X <- sim$states$x$i0

# Grab all the simulations (the first five columns are indices, time, etc.)
Y <- X[,-c(1:5)]

# Grab prediction time column
t <- X[,"t.j"]

# Plot
matplot(t,Y,type="l", ylim=c(-4,4))

Lets see the effect of setting $\sigma_{x} = 3$:

sim <- model$simulate(data=.data, 
                      pars=c(20,3,0.05), 
                      n.sims=100,
                      silent=T)

Lets see the effect of setting $\theta = 50$:

sim <- model$simulate(data=.data, 
                      pars=c(50,1,0.05), 
                      n.sims=100,
                      silent=T)