tests/example.R

In svdataman/gin: Gaussian Processes for time series inference

# define an ACV function
acv <- function(theta, tau) {
  A <- abs(theta[1])
  l <- abs(theta[2])
  acov <- A * exp(-(tau / l))
  return(acov)
}

# define parameters of ACV
# theta[1] = mu (mean)
# theta[2] = nu (error scale factor)
# theta[3:p] = parameters of ACV
theta <- c(0.0, 1.0, 1.0, 1.0)

# define vector of times for observations
m <- 1000
t <- seq(-0.5, 10.5, length = m)

# produce Gaussian vector (simulation)
y <- gp_sim(theta, acv.model = acv, t.star = t)
y <- as.vector(y)

# plot the 'true' curve
dat.true <- data.frame(t = t, y = y)
plot_ts(dat.true, type = "l")

# --------------------------------------------
# produce Gaussian vector (simulation)
n <- 5
y <- gp_sim(theta, acv.model = acv, t.star = t, n.sim = n)

for (i in 1:n) {
  dat <- data.frame(t = t, y = y[, i])
  plot_ts(dat, col = i, add = TRUE, type = "l")
}
# --------------------------------------------
# examine some data

plot_ts(drw, col = "black", cex.lab=1.4)

# define an ACV function
acv <- function(theta, tau) {
  A <- abs(theta[1])
  l <- abs(theta[2])
  acov <- A * exp(-(tau / l))
  return(acov)
}

# define parameters of ACV
# theta[1] = mu (mean)
# theta[2] = nu (error scale factor)
# theta[3:p] = parameters of ACV
theta <- c(0.0, 1.0, 1.0, 1.0)

# fit the model to the data, find Max.Like parameter values
result <- gp_fit(theta, acv.model = acv, dat = drw)

# define vector of times for reconstruction
m <- 1000
t <- seq(-0.5, 10.5, length = m)

# reconstruct process: compute 'conditional' mean and covariance
gp <- gp_conditional(result$par, acv.model = acv, drw, t.star = t)

# plot a 'snake' showing mean +/- std.dev
plot_snake(gp, add = TRUE, col.line = 3)

# --------------------------------------------
# produce Gaussian vector (simulation)

y.sim <- gp_sim(result$par, dat = drw, acv.model = acv, t.star = t,
            N.sim = 5, plot = FALSE)

for (i in 1:5) lines(t, y.sim[, i], col = i)