convert_sequence_format() turns wide or long sequence data into one of
four data frames: per-sequence state counts, per-sequence state presence
(one-hot), the list of consecutive transitions, or each event paired with
the one before it. It reads an event log (action set) or a wide table
(one row per sequence, one column per position). The counts can serve as
per-sequence behaviour profiles for the association estimators of
build_network() and for build_mcml_pc(). The id columns (those that
id_col names, and .rid for a wide table without id_col) are not
variables and must be set aside there (id_col in build_mcml_pc(),
actor in build_network()); the rid column is
dropped automatically.
Usage
convert_sequence_format(
data,
seq_cols = NULL,
id_col = NULL,
action = NULL,
time = NULL,
format = c("frequency", "onehot", "edgelist", "follows")
)Arguments
- data
A data frame of sequences: an event log or a wide table.
- seq_cols
The state columns of a wide table; setting them declares a wide table. Default
NULL: every column exceptid_colandtime. With neitheractionnorseq_cols, the frame must be one column or hold position columns named as a numbered run (T1,T2, ..., or1,2, ...), with nothing beside them but theid_col(andtime) columns; anything else raises an error of classnestimate_bad_input, as does a column listed both here and inid_colortime.- id_col
The identifier column(s), copied to every output row. In an event log, each unique combination of their values is one sequence, and it is required: an event log without
id_colraises an error of classnestimate_bad_input(no column is taken by position). In a wide table, each row is one sequence and no id is inferred from the data: withoutid_col, a row-index column.ridis added. DefaultNULL.- action
The column holding each event's state; setting it declares an event log. Default
NULL.- time
The column that orders the events within each sequence of an event log; without it, events keep their row order. A wide table leaves it out of the state columns. Default
NULL.- format
The output format:
"frequency"(default), the count of each state per sequence;"onehot", whether each state occurs in the sequence (0/1);"edgelist", each pair of consecutive states (from,to);"follows", each event (act) paired with the event before it (follows).
Value
A data frame with the id column(s) first. For "frequency" and
"onehot": one row per sequence, a rid column identifying the
sequence, and one integer column per state, holding counts or 0/1. For
"edgelist": one row per transition, with columns from and to. For
"follows": one row per event after the first of its sequence, with
columns act and follows. An event log's rows come out sorted by
id_col.
Details
A missing state (NA or an empty string) is dropped and the states on
either side are joined, so a sequence A, NA, B gives the transition
A -> B. frequencies() counts no transition across a missing state
instead.
See also
frequencies for building transition frequency matrices.
Examples
# Wide format input
seqs <- data.frame(V1 = c("A","B","A"), V2 = c("B","A","C"), V3 = c("A","C","B"))
convert_sequence_format(seqs, format = "frequency")
#> .rid rid A B C
#> 1 1 1 2 1 0
#> 2 2 2 1 1 1
#> 3 3 3 1 1 1
convert_sequence_format(seqs, format = "edgelist")
#> .rid from to
#> 1 1 A B
#> 2 1 B A
#> 3 2 B A
#> 4 2 A C
#> 5 3 A C
#> 6 3 C B