mpadge
commented
3 years ago The second of the above commits replaces the non-transparent cutoff parameter with a much more transparent prop_stops parameter, which enables the timetable scan to stop once the specified proportion of all stations have been reached. This is the result:
library (gtfsrouter)
gtfs <- extract_gtfs("vbb.zip")
#> ▶ Unzipping GTFS archive
#> ✔ Unzipped GTFS archive
#> ▶ Extracting GTFS feed✔ Extracted GTFS feed
#> ▶ Converting stop times to seconds✔ Converted stop times to seconds
#> ▶ Converting transfer times to seconds✔ Converted transfer times to seconds
transfers <- gtfsrouter::gtfs_transfer_table (gtfs, network_times = FALSE)
transfers <- gtfs$transfers %>%
select(.data$from_stop_id, .data$to_stop_id, .data$transfer_type, .data$min_transfer_time) %>%
rbind(transfers) %>%
group_by(.data$from_stop_id, .data$to_stop_id) %>%
summarise(across(everything(), first)) %>% # take min_transfer_time from gtfs$transfers if present
data.table::data.table()
gtfs$transfers <- transfers
gtfs <- gtfs_timetable(gtfs, day = "tuesday")
from <- "Berlin Hauptbahnhof"
start_time <- 8 * 3600
p <- c (5:9 / 10, 0.95, 0.99, 1) # values for new prop_stops parameter
res <- lapply (p, function (i) {
st <- system.time (
x <- gtfs_traveltimes (gtfs, from = from, start_time = start_time, prop_stops = i)
)
c (nrow (x), max (x$duration), st [3]) })
res <- do.call (rbind, res)
dat <- data.frame (prop_stops = p,
n_stations = res [, 1] / 1000,
max_duration = res [, 2] / 3600,
time = res [, 3])
scl <- max (dat$time) / as.integer (max (dat$max_duration))
library (ggplot2)
ggplot(dat, aes (x = prop_stops)) +
geom_point (aes (y = max_duration, colour = "max_duration"), size = 2) +
geom_line (aes (y = max_duration, linetype = "max_duration", colour = "max_duration"), size=1.1) +
geom_line (aes (y = time / scl, linetype = "time", colour = "time"), size=1.1) +
geom_point (aes (y = time / scl, colour = "time"), size = 2) +
scale_y_continuous (sec.axis = sec_axis (~.*scl, name = "Calculation time (seconds)")) +
labs (y = "Max. duration (hours)") +
scale_linetype_manual ("colour", values = c ("time" = "solid", "max_duration" = "solid")) +
theme (legend.title = element_blank (),
legend.background = element_blank (),
legend.position = c (0.1, 0.9))
Created on 2021-02-01 by the reprex package (v1.0.0)
We could easily expose a function like this that could be used to help determine an appropriate value for prop_stops for a given feed. In the case here for the VBB feed, the value would look like this:
imax <- which.max (diff (max_duration)) - 1L
dat$prop_stops [imax]
#>[1] 0.9With that value applied to the data above giving:
system.time (
x <- gtfs_traveltimes (gtfs, from = from, start_time = start_time, prop_stops = 0.9)
)
#> user system elapsed
#> 3.684 0.016 3.632
format (nrow (x), big.mark = ",")
#> [1] "35,870"Created on 2021-02-01 by the reprex package (v1.0.0)
So that takes quite a bit longer than the default value for the previous cutoff parameter, but that one (with cutoff = 0.1) only reached 13,711 stations, whereas this one reaches almost 3 times as many stations. Times basically scale linearly with prop_stops up to around 0.95 or so, with the previous cutoff = 0.1 parameter corresponding to prop_stops = 0.5. This new prop_stops parameter is, I think, much clearer and enables much more direct control over what is being returned.
AlexandraKapp
Following on from #57 and comment by @AlexandraKapp, setting
cutoff = 0now takes way too long, and moreover produces this result:Created on 2021-02-01 by the reprex package (v1.0.0)
So the full scan (
cutoff = 0) is scanning way too many lines of the timetable. (The number reached can be more than the total number reached in the timetable because of transfers to stations that aren't otherwise reachable from the timetable itself.) First thing to do will be to implement a stop clause that stops scanning the timetable as soon as all possible stations have been reached.