build_atna() estimates an attention-weighted transition network (ATNA):
every later state in a sequence, not only the next one, adds to the edge
from an earlier state, with a weight that decays with the number of
positions between them, exp(-distance / lambda). It is
build_network() with method = "attention". The decay is set in
params: lambda (default 1; larger values decay more slowly) and
direction ("forward", the default, "backward" or "both").
duration (one value per position, accumulated into times) or
time_matrix (one row per sequence, one column per position) measure
the distance in time instead of positions, and decay, a function of
(ti, tj, lambda) returning one weight per pair, replaces the decay
function.
The weights are summed decay weights, not counts or probabilities, so they
compare within one network; add scaling = "normalize" for rows that sum
to 1. Check how far they move under resampling with
bootstrap_network().
Usage
build_atna(
data,
start = FALSE,
end = FALSE,
actor = NULL,
action = NULL,
time = NULL,
session = NULL,
...
)Arguments
- data
The sequences: a data frame with one row per sequence and one column per position, or an event log (one row per event, with
actionset). Column roles are declared as inbuild_network(). One-hot data and matrices are refused.- start
A start state put before every sequence, so the network shows where sequences begin.
FALSE(default) adds none,TRUEadds one labelled"Start", a single string adds one with that label.- end
An end state put after each sequence's last observed state, so the network shows where sequences stop; it has no outgoing edges.
FALSE(default),TRUE("End") or a single string, asstart.- actor
The column naming who each row belongs to (a person, a team). In an event log, each actor's events form a sequence, split further by
sessionortime_threshold. DefaultNULL.- action
The column holding each event's state; setting it declares the data an event log. Column names give no role:
action,actor,timeandsessionare read only when set. A column calledActionwith nothing declared raises an error of classnestimate_bad_input, and an undeclared time or session column besideactionis named in a message of classnestimate_column_role. DefaultNULL.- time
The timestamp column of an event log. Events are ordered by it, and a gap longer than
time_thresholdstarts a new sequence. DefaultNULL.- session
The column identifying sessions in an event log; each session is its own sequence, also without
actor. DefaultNULL.- ...
Further arguments for
build_network(), such asgroup,params,scaling,thresholdorstate_cols.methodcannot be given: it is always"attention"(error of classnestimate_bad_input).
Value
A netobject (a netobject_group with group), exactly as
build_network() returns it: print it for an overview, read the edges
with extract_edges() and the weights with
extract_transition_matrix(), and draw it with cograph::splot().
Examples
net <- build_atna(trajectories)
net
#> Attention Network (decay-weighted transitions) [directed]
#> Weights: [15.638, 251.913] | mean: 107.233
#>
#> Weight matrix:
#> Active Average Disengaged
#> Active 251.098 101.100 15.638
#> Average 89.399 251.913 76.872
#> Disengaged 22.296 71.697 85.083
#>
#> Initial probabilities:
#> Active 0.382 ████████████████████████████████████████
#> Average 0.368 ██████████████████████████████████████
#> Disengaged 0.250 ██████████████████████████