Construct a transition network analysis (TNA) model from sequence data. The function takes a data set of sequence of events or states as input and builds a TNA model. It extracts the edge weights and initial probabilities from the data along with the state labels. The function also accepts weight matrices and initial state probabilities directly.
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)# S3 method for default
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for matrix
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for stslist
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for data.frame
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for tna_data
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for tsn
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
# S3 method for tsn_ews
build_model(
x,
type = "relative",
scaling = character(0L),
cols = tidyselect::everything(),
params = list(),
inits,
begin_state,
end_state
)
tna(x, ...)
ftna(x, ...)
ctna(x, ...)
atna(x, ...)
tsn(x, ...)
An object of class tna which is a list containing the
following elements:
weights: An adjacency matrix of the model (weight matrix).
inits: A numeric vector of initial values for each state.
For matrix type x, this element will be NULL if inits is not
directly provided
labels: A character vector of the state labels, or NULL if
there are no labels.
data: The original sequence data that has been converted to an
internal format used by the package when x is a stslist or a
data.frame object. Otherwise NULL.
A stslist (from TraMineR), data.frame, a matrix, or
a tna_data object (see prepare_data()).
For stslist and data.frame objects x
should describe a sequence of events or states to be used for building the
Markov model. If x is a matrix, it is assumed that the element on row
i and column j is the weight of the edge representing the transition
from state i to state j. If x is a data.frame, then
it must be in wide format (see cols on how to define columns for the
time points).
A character string describing the weight matrix type.
Currently supports the following types:
"relative" for relative frequencies (probabilities, the default)
"frequency" for frequencies.
"co-occurrence" for co-occurrences.
"n-gram" for n-gram transitions. Captures higher-order transitions by
considering sequences of n states, useful for identifying longer
patterns.
"gap" allows transitions between non-adjacent states, with
transitions weighted by the gap size.
"window" creates transitions between all states within a
sliding window, capturing local relationships
(several sequences together).
"reverse" considers the sequences in reverse order
(resulting in what is called a reply network in some contexts).
The resulting weight matrix is the transpose of the "frequency"
option.
"attention" aggregates all downstream pairs of states with an
exponential decay for the gap between states. The parameter lambda
can be used to control the decay rate (the default is 1)-
A character vector describing how to scale the weights
defined by type. When a vector is provided, the scaling options are
applied in the respective order. For example, c("rank", "minmax") would
first compute the ranks, then scale them to the unit interval using
min-max normalization. An empty vector corresponds to no scaling.
Currently supports the following options:
"minmax" performs min-max normalization to scale the weights to the
unit interval. Note that if the smallest weight is positive, it
will be zero after scaling.
"max" Multiplies the weights by the reciprocal of the largest weight
to scale the weights to the unit interval. This options preserves
positive ranks, unlike "minmax" when all weights are positive.
"rank" Computes the ranks of the weights using base::rank() with
ties.method = "average".
An expression giving a tidy selection of columns that should
be considered as sequence data. By default, all columns are used. Ignored
for matrix, tna_data and tsn type x.
A list of additional arguments for models of specific
type. The potential elements of this list are:
n_gram: An integer for n-gram transitions specifying the number of
adjacent events. The default value is 2.
max_gap: An integer for the gap-allowed transitions specifying the
largest allowed gap size. The default is 1.
window_size: An integer for the sliding window transitions
specifying the window size. The default is 2.
weighted: A logical value. If TRUE, the transitions
are weighted by the inverse of the sequence length. Can be used for
frequency, co-occurrence and reverse model types. The default is
FALSE.
direction: A character string specifying the direction of attention
for models of type = "attention". The available options are
"backward", "forward", and "both", for backward attention,
forward attention, and bidirectional attention, respectively.
The default is "forward".
decay: A function that specifies the decay of the weights between
two time points at a specific distance. The function should take three
arguments: i, j and lambda, where i and j are numeric
vectors of time values, and lambda is a numeric value for the
decay rate. The function should return a numeric vector of weights.
The default is function(i, j, lambda) exp(-abs(i - j) / lambda).
lambda: A numeric value for the decay rate. The default is 1.
time: A matrix or a data.frame providing the time values
for each sequence and at time index. For tna_data objects, this can
also be a logical value, where TRUE will use the time_data element
of x for the time values. Date values are converted to numeric.
duration: A matrix or a data.frame providing the
time spent in each state for each sequence and time index.
This is an alternative to time.
An optional numeric vector of initial state probabilities
for each state. The vector will be scaled to unity.
Ignored if x is not a matrix.
A character string for an additional begin state.
This state is added as the first observation for every sequence to
signify the beginning of the sequence
A character string for an additional end state.
This state is added as the last observation for every sequence to
siginify the end of the sequence.
Ignored. For the build_model aliases (e.g., tna), this
argument matches the actual arguments to build_model beside x.
Basic functions
hist.group_tna(),
hist.tna(),
plot.group_tna(),
plot.tna(),
plot_frequencies(),
plot_frequencies.group_tna(),
plot_mosaic(),
plot_mosaic.group_tna(),
plot_mosaic.tna_data(),
print.group_tna(),
print.summary.group_tna(),
print.summary.tna(),
print.tna(),
summary.group_tna(),
summary.tna(),
tna-package
model <- build_model(group_regulation)
print(model)
model <- tna(group_regulation)
model <- ftna(group_regulation)
model <- ctna(group_regulation)
model <- atna(group_regulation)
Run the code above in your browser using DataLab