Behaves like subsetting a data frame, with three additions. If the result
no longer carries the columns that make a journey table a journey table,
the mm_paths class is dropped and a plain data frame is returned, so a
column subset cannot produce an object that claims to be an mm_paths but
cannot answer any question about journeys. Rows selected by a missing
(NA) row index are dropped with a message, as subset() drops them. And
touch_rank and touch_n are recomputed whenever they no longer match the
rows present.
Details
When a row subset removes part of a journey, dropping one channel say,
the first remaining touch becomes rank 1 and the result is a valid journey
table in its own right, so every credit rule sums to 1 per converting
journey. A row selected twice counts as two touches. Construction counts
describe the table build_paths() returned, so they are dropped once rows
change. Convert with as.data.frame() first to inspect the original ranks
of a partial selection.
The dplyr verbs follow the same rules: filter(), slice(),
arrange() and distinct() subset rows as [ does, and mutate() and
the joins recompute ranks the same way. Journey tables built by separate
build_paths() calls share journey ids; combine them with bind_paths().
Examples
data(mm_events)
paths <- build_paths(mm_events, id = "customer_id", channel = "channel",
timestamp = "timestamp", conversion = "conversion")
# Row subset: still a journey table
class(paths[paths$converted, ])
#> [1] "mm_paths" "data.frame"
# Column subset that drops the structure: plain data frame
class(paths[, c("channel", "touch_rank")])
#> [1] "data.frame"