In this vignette we present additional features of the
sfclust package.
We begin by loading the required packages.
The simulated dataset used in this vignette, stgaus, is
included in our package. It is a stars object with one
variable, y, and two dimensions: geometry and
time. The dataset represents continuous Gaussian
measurements from 100 regions, observed daily over 91 days, starting in
January 2024.
#> stars object with 2 dimensions and 1 attribute
#> attribute(s):
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> y -1.170275 -0.2390084 0.0302046 -0.0004274323 0.2095926 1.370593
#> dimension(s):
#> from to offset delta refsys point
#> geometry 1 100 NA NA NA FALSE
#> time 1 91 2024-01-01 1 days Date FALSE
#> values
#> geometry POLYGON ((59.5033 683.285...,...,POLYGON ((942.7562 116.89...
#> time NULL
To gain some initial insights, we can aggregate the data weekly:
stweekly <- aggregate(stgaus, by = "week", FUN = mean)
ggplot() +
geom_stars(aes(fill = y), stweekly) +
facet_wrap(~ time, ncol = 5) +
scale_fill_distiller(palette = "RdBu") +
theme_bw()We can also examine the trends for each region:
stgaus |>
st_set_dimensions("geometry", values = 1:nrow(stgaus)) |>
as_tibble() |>
ggplot() +
geom_line(aes(time, y, group = geometry, color = factor(geometry)), linewidth = 0.3) +
theme_bw() +
theme(legend.position = "none")Some regions exhibit similar trends over time, but the overall patterns are more complex than polynomial functions. Our goal is to cluster these regions while accounting for spatial contiguity.
We assume that the response variable \(Y_{it}\) for region \(i\) at time \(t\) follows a normal distribution given the partition \(M\): \[ Y_{it} \mid \mu_{it}, M \stackrel{ind}{\sim} \text{Normal}(\mu_{it}, \sigma^2), \] where the mean \(\mu_{it}\) is modeled based on the cluster assignment \(c_i\): \[ \mu_{it} = \alpha_{c_i} + f_{c_i,t}, \] where \(\alpha_{c_i}\) is the intercept for cluster \(c_i\), and \(f_{c_i,t}\) is a latent random effect modeled as a random walk process. Specifically, we impose the condition: \[ f_{c_i,t} - f_{c_i,t-1} \sim N(0, \nu^{-1}), \quad \text{for } t = 2, \dots, n. \]
The prior for the hyperparameter \(\nu_c\) is defined as: \[ \log(\nu_c) \sim \text{Normal}(-2, 1). \]
sfclust uses an undirected graph to represent
connections between regions and proposes spatial clusters using this
graph through minimum spanning trees (MST). By default,
sfclust accepts the argument graphdata, which
should include:
igraph object representing spatial
connections,membership vector indicating the initial cluster
assignments for each region.For simplicity, you can use the genclust function to
generate an initial random partitioning with a specified number of
clusters. In this example, we create a partition with 20 clusters:
#> [1] "graph" "mst" "membership" "valid_ids"
Now, let’s visualize how the regions were randomly clustered. Panel (A) shows the full adjacency graph (all neighbor connections), panel (B) shows the MST derived from it (the backbone used for cluster proposals), and panel (C) shows the resulting initial partition:
gg1 <- ggraph(initial_cluster$graph, layout = st_coordinates(st_centroid(st_geometry(stgaus)))) +
geom_edge_fan(linetype = 1, color = 2) +
geom_node_point(size = 1.5, color = 1) +
geom_sf(data = st_geometry(stgaus), fill = NA, color = 1, linewidth = 0.5) +
labs(subtitle = "(A)") +
theme_void()
gg2 <- ggraph(initial_cluster$mst, layout = st_coordinates(st_centroid(st_geometry(stgaus)))) +
geom_edge_fan(linetype = 1, color = 2) +
geom_node_point(size = 1.5, color = 1) +
geom_sf(data = st_geometry(stgaus), fill = NA, color = 1, linewidth = 0.5) +
labs(subtitle = "(B)") +
theme_void()
gg3 <- st_sf(st_geometry(stgaus), cluster = factor(initial_cluster$membership)) |>
ggplot() +
geom_sf(aes(fill = cluster), color = 1) +
labs(subtitle = "(C)") +
theme_void() +
theme(legend.position = "none")
gg1 + gg2 + gg3 & theme(plot.margin = margin(0, 0, 0, 0))sfclustWe will initiate the Bayesian clustering algorithm using our
generated partition (initial_cluster). Additionally, we use
the capabilites of INLA::inla to define a model that
includes a random walk process in the linear predictor, and custom
priors for the scale hyperparameter. To begin, we will run the algorithm
for only 50 iterations.
The logpen argument sets the log-penalty on the number
of clusters, specifically log(1-q) where q is
the prior probability of keeping a cluster. A large negative value such
as -50 favors fewer clusters, acting as a strong parsimony
prior. The burnin argument discards the first iterations
before saving samples, and thin retains only every
thin-th iteration, reducing autocorrelation in the
chain.
result0 <- sfclust(stgaus, graphdata = initial_cluster, logpen = -50,
formula = y ~ f(id_time, model = "rw1",
hyper = list(prec = list(prior = "normal", param = c(-2, 1)))),
niter = 50, burnin = 10, thin = 2, nmessage = 10
)
result0#> Within-cluster formula:
#> y ~ f(id_time, model = "rw1", hyper = list(prec = list(prior = "normal",
#> param = c(-2, 1))))
#>
#> Clustering hyperparameters:
#> log(1-q) birth death change hyper
#> -50.000 0.425 0.425 0.100 0.050
#>
#> Clustering movement counts:
#> births deaths changes hypers
#> 11 8 1 2
#>
#> Log marginal likelihood (sample 25 out of 25): 2570.063
The output indicates that after starting with 20 clusters, the algorithm created 11 new clusters (births), removed 8 clusters (deaths), changed the membership of 1 cluster, and modified the minimum spanning tree (MST) 2 times.
The plot method allows us to select which graph to
produce. For example, we can visualize only the log marginal likelihood
to diagnose convergence. With just 50 iterations, we can see that the
log marginal likelihood has not yet achieved convergence.
To achieve convergence, we can continue the sampling using the
update function and specify the number of new iterations
with the niter argument:
#> Within-cluster formula:
#> y ~ f(id_time, model = "rw1", hyper = list(prec = list(prior = "normal",
#> param = c(-2, 1))))
#>
#> Clustering hyperparameters:
#> log(1-q) birth death change hyper
#> -50.000 0.425 0.425 0.100 0.050
#>
#> Clustering movement counts:
#> births deaths changes hypers
#> 33 44 7 43
#>
#> Log marginal likelihood (sample 1000 out of 1000): 12602.37
With 1000 additional iterations, there have been many clustering movements. Furthermore, when visualizing the results, we can observe that the log marginal likelihood has achieved convergence.
The side-by-side comparison below highlights the contrast: panel (A) shows the unstable log marginal likelihood from the short initial run, while panel (B) shows the stabilized trajectory after the full run, indicating convergence.
gg1 <- plot(result0, which = 3) + labs(subtitle = "(A)")
gg2 <- plot(result, which = 3) + labs(subtitle = "(B)")
gg1 + gg2if (save_figures) {
ggsave(file.path(path_figures, "stgaus-marginal-likelihood.pdf"), width = 10, height = 4.5,
device = cairo_pdf)
}The update function can also be used to select a
different past iteration as the active clustering, without running
additional MCMC. This is useful when you want to inspect or use a
specific sample rather than the last one:
#> Within-cluster formula:
#> y ~ f(id_time, model = "rw1", hyper = list(prec = list(prior = "normal",
#> param = c(-2, 1))))
#>
#> Clustering hyperparameters:
#> log(1-q) birth death change hyper
#> -50.000 0.425 0.425 0.100 0.050
#>
#> Clustering movement counts:
#> births deaths changes hypers
#> 33 44 7 43
#>
#> Log marginal likelihood (sample 750 out of 1000): 12319.21
The final iteration indicates that the algorithm identified 12 clusters, with a log marginal likelihood of 12,602.37. The largest cluster consists of 27 regions, while the smallest contains only one region.
#> Summary for clustering sample 1000 out of 1000
#>
#> Within-cluster formula:
#> y ~ f(id_time, model = "rw1", hyper = list(prec = list(prior = "normal",
#> param = c(-2, 1))))
#>
#> Counts per cluster:
#> 1 2 3 4 5 6 7 8 9 10 11 12
#> 27 20 12 11 9 8 6 3 1 1 1 1
#>
#> Log marginal likelihood: 12602.37
Let’s visualize the regions grouped by cluster.
Let’s visualize the original data grouped by cluster.
plot_clusters_series(result, y, sort = TRUE) +
facet_wrap(~ cluster, ncol = 5) +
labs(title = "Risk per cluster", y = "Response")