# NOT RUN {
# As an example, create a simulated SSN object
# Save the object to a temporary location
set.seed(12)
ssn_path <- paste(tempdir(), "/example_network", sep = "")
# If example network doesn't already exist, then attempt to create it
# Otherwise, read from the temporary directory
example_network <- try(importSSN(ssn_path, predpts = 'preds', o.write = TRUE), silent = TRUE)
if('try-error' %in% class(example_network)){
example_network <- createSSN(
n = 50,
obsDesign = binomialDesign(200),
predDesign = binomialDesign(50),
importToR = TRUE,
path = ssn_path,
treeFunction = iterativeTreeLayout
)
}
# plot the simulated network structure with prediction locations
# plot(example_network, bty = "n", xlab = "x-coord", ylab? = "y-coord")
## create distance matrices, including between predicted and observed
createDistMat(example_network, "preds", o.write = TRUE, amongpred = TRUE)
## extract the observed and predicted data frames
observed_data <- getSSNdata.frame(example_network, "Obs")
prediction_data <- getSSNdata.frame(example_network, "preds")
## associate continuous covariates with the observation locations
# data generated from a normal distribution
obs <- rnorm(200)
observed_data[,"X"] <- obs
observed_data[,"X2"] <- obs^2
## associate continuous covariates with the prediction locations
# data generated from a normal distribution
pred <- rnorm(50)
prediction_data[,"X"] <- pred
prediction_data[,"X2"] <- pred^2
## simulate some Gaussian data that follows a 'tail-up' spatial process
sims <- SimulateOnSSN(
ssn.object = example_network,
ObsSimDF = observed_data,
PredSimDF = prediction_data,
PredID = "preds",
formula = ~ 1 + X,
coefficients = c(1, 10),
CorModels = c("Exponential.tailup"),
use.nugget = TRUE,
CorParms = c(10, 5, 0.1),
addfunccol = "addfunccol")$ssn.object
## extract the observed and predicted data frames, now with simulated values
sim1DFpred <- getSSNdata.frame(sims, "preds")
sim1preds <- sim1DFpred[,"Sim_Values"]
sim1DFpred[,"Sim_Values"] <- NA
sims <- putSSNdata.frame(sim1DFpred, sims, "preds")
# create the adjacency matrix for use with smnet
adjacency <- get_adjacency(
ssn_path,
net = 1
)
# not run - plot the adjacency matrix
# display(adjacency[[1]])
# sometimes it is useful to see which variables are valid network weights
# in the data contained within the SSN object
show_weights(sims, adjacency)
# fit a penalised spatial model to the stream network data
# Sim_Values are quadratic in the X covariate. To highlight
# the fitting of smooth terms, this is treated as non-linear
# and unknown using m().
mod_smn <- smnet(formula = Sim_Values ~ m(X) + m(X2) +
network(adjacency = adjacency, weight = "shreve"),
data.object = sims, netID = 1)
# not run - plot different summaries of the model
plot(mod_smn, type = "network-covariates")
plot(mod_smn, type = "network-segments", weight = 4, shadow = 2)
plot(mod_smn, type = "network-full", weight = 4, shadow = 2)
# obtain predictions at the prediction locations and plot
# against true values
preds <- predict(mod_smn, newdata = getSSNdata.frame(sims, "preds"))
plot(preds$predictions, sim1preds)
# obtain summary of the fitted model
summary(mod_smn)
# delete the simulated data
unlink(ssn_path, recursive = TRUE)
# }
Run the code above in your browser using DataLab