macro_network() returns the cluster-level (macro) network of an mcml,
optionally with one or more clusters expanded back into their member
states. Every other cluster stays collapsed to a single node, so the
result is a network at mixed resolution: the cluster of interest in
detail, its context in summary, with weights counted from the sequences.
For an mcml_pc from build_mcml_pc(), it returns the cluster-level
network that fit already holds.
Usage
macro_network(
x,
expand = NULL,
method = "relative",
scaling = NULL,
threshold = 0,
params = list(),
...
)Arguments
- x
An
mcmlbuilt from sequence data, or anmcml_pcfrombuild_mcml_pc(). Anmcmlbuilt from a matrix or an edge list holds no sequences to count, so it raises an error of classnestimate_no_expand_source, even withexpand = NULL; a name inexpandthat is not a cluster is reported first.- expand
The clusters to expand into their member states.
NULL(default) collapses every cluster, which reproduces the macro layer;"all"orTRUEexpands every cluster, so each state is its own node. A name that is not a cluster raises an error. For anmcml_pc, any value other thanNULLraises an error of classnestimate_no_expand: there are no sequences to count again.- method
The estimator passed to
build_network(). Default"relative"(row-normalised transitions). For anmcml_pc, giving it raises an error: the estimator was set bybuild_mcml_pc().- scaling, threshold, params
Passed to
build_network()for the re-counted network:scaling(defaultNULL, none),threshold(default0) andparams(defaultlist()), as there. For anmcml_pc, giving any raises an error, as formethod.- ...
Further arguments for
build_network(), such asstartorend. For anmcml_pc, giving any raises an error, as formethod.
Value
For an mcml: a netobject (also a cograph_network) whose
nodes are the collapsed clusters plus the member states of any
expanded cluster, with weights counted by build_network(). It is
printed, resampled (for example with bootstrap_network()) and drawn
like any other netobject; its node grouping is read by cograph's
group-aware plots, such as cograph::plot_mtna(). Beside the usual
netobject fields it holds node_groups, a data frame with the
columns node and group that maps each node to its cluster (an
expanded cluster's states map to that cluster, a collapsed cluster maps
to itself), the same labels as a factor in the groups column of
nodes, and
expanded, the names of the clusters expanded (NULL when none
were).
For an mcml_pc: its cluster-level network, the netobject estimated
by build_mcml_pc(), unchanged.
Details
The weights are counted again from the sequences, after each state is recoded to its cluster (or kept, for an expanded cluster), because a cluster-level matrix cannot be split back into its states.
Examples
seqs <- data.frame(
t1 = c("A", "C", "A", "B"), t2 = c("B", "D", "C", "A"),
t3 = c("C", "A", "D", "C"), stringsAsFactors = FALSE
)
mc <- build_mcml(seqs, clusters = list(G1 = c("A", "B"), G2 = c("C", "D")))
macro_network(mc) # every cluster collapsed
#> Transition Network (relative probabilities) [directed]
#> Weights: [0.333, 0.667] | mean: 0.500
#>
#> Weight matrix:
#> G1 G2
#> G1 0.400 0.600
#> G2 0.333 0.667
#>
#> Initial probabilities:
#> G1 0.750 ████████████████████████████████████████
#> G2 0.250 █████████████
macro_network(mc, expand = "G2") # G2 shown as C and D
#> Transition Network (relative probabilities) [directed]
#> Weights: [0.400, 1.000] | mean: 0.750
#>
#> Weight matrix:
#> C D G1
#> C 0.0 1 0.0
#> D 0.0 0 1.0
#> G1 0.6 0 0.4
#>
#> Initial probabilities:
#> G1 0.750 ████████████████████████████████████████
#> C 0.250 █████████████
#> D 0.000