frequencies.txt exact_times=1 trip_id semantics

[derhuerst]

I have a question about the semantics of frequencies.txt with exact_times=1.

---

From my experience with GTFS and my observations of GTFS-based UIs, a lot of tools seem to make the following assumption:

A single trip (as defined by a unique trip_id in the GTFS dataset) is one vehicle (or group of vehicles) that I can use without significant interruptions (such as waiting for another vehicle or chasing to a different line). Often, a single trip is considered to be a vehicle which I can travel with for the whole duration of the trip.

The frequencies.txt documentation seems to undermine that assumption however:

frequencies.txt represents trips that operate on regular headways (time between trips). This file can be used to represent two different types of service.

• Frequency-based service (exact_times=0) in which service does not follow a fixed schedule throughout the day. Instead, operators attempt to strictly maintain predetermined headways for trips.
• A compressed representation of schedule-based service (exact_times=1) that has the exact same headway for trips over specified time period(s). In schedule-based service operators try to strictly adhere to a schedule.

Field Name	Type	Required	Description
trip_id	ID referencing trips.trip_id	Required	Identifies a trip to which the specified headway of service applies.

I think this is especially important for routing engines: Now, they can't assume anymore that every GTFS data point referring to the same trip_id is tied to one vehicle allowing continuous travel. There are >=1 "runs" of a vehicle, all under the same trip_id, but each of them ends at the last stop specified in stop_times.txt.

---

Is my understanding of the semantics correct? If it is, I'd argue that this is quite unintuitive and therefore easy to implement in a wrong way. If I misunderstood how frequencies.txt works, let's improve the documentation.

[skinkie]

On Saturday, June 6, 2020 8:05:27 PM CEST, Jannis Redmann wrote:

Is my understanding of the semantics correct? If it is, I'd argue that this is quite unintuitive and therefore easy to implement in a wrong way. If I misunderstood how frequencies.txt works, let's improve the documentation.

Could you give an example how it could be interpreted in a wrong way?

[derhuerst]

Let's consider an excerpt from the example feed linked in the spec:

trip_id	arrival_time	departure_time	stop_id	stop_sequence	pickup_type	drop_off_type
AWE1	0:06:10	0:06:10	S1	1	0	0
AWE1			S2	2	1	3
AWE1	0:06:20	0:06:30	S3	3	0	0
AWE1			S5	4	0	0
AWE1	0:06:45	0:06:45	S6	5	0	0

trip_id	start_time	end_time	headway_secs
AWE1	05:30:00	06:30:00	300

If I assume that all stop_times.txt/frequencies.txt for AWE1 describe one "run" of one vehicle that I can use continuously, then I could conclude that I can stay in the vehicle from 05:30:00 (earliest start in time frame) until 7:00:00 (latest start in time frame + 35min). This is not the case I assume?

[skinkie]

It is not one run (or block). It is a normalisation form of transit data. Including a confidence interval of the arrival time of the next trip. Remaining in the vehicle for no particular reason is an activity that is probably allowed if you would have a day ticket, but that is not what this structure (or GTFS) explicitly defines.

[derhuerst]

It is not one run (or block).

(Not sure what exactly you mean by "run" here, but I will assume you mean what I tried to explain.)

In the GTFS ecosystem, I have often observed the assumption that one GTFS trip corresponds to exactly one "run". Or in plain English: That one GTFS trip means that one vehicle will continuously visit all stops in the trip, without any other trips in between and without additional stops before or after; That after the vehicle has visited all stops in the trip, the "run" is "over".

Making that assumption would probably lead to routing errors (e.g. routes that I actually can't take or that are physically impossible) & unintuitive UIs (e.g. showing the first stop of the trip between other later stops, because another "run" in a time frame of compressed data has started).

If this assumption is not to be made, meaning the stop_times/frequencies feature of GTFS is purely a "normalisation form" to describe when & where any appropriate vehicle of a line will stop, IMO we should clarify this better in the documentation.

(All of this does of course not apply anyways to different schemes of sending vehicles around, like circle-based lines or lines split up by direction.)

[antrim]

GTFS Best Practices offer the below.

Field Name	Recommendation
block_id	Can be provided for frequency-based trips.

So, that means the following example is valid, and indicates a continuous loop where passengers can stay onboard at stop_A.

stop_times.txt

trip_id	arrival_time	departure_time	stop_id	stop_sequence
trip_1	06:10:00	06:10:00	stop_A	1
trip_1	06:15:00	06:15:00	stop_B	2
trip_1	06:20:00	06:20:00	stop_C	3
trip_1	06:25:00	06:25:00	stop_D	4
trip_1	06:30:00	06:30:00	stop_E	5
trip_1	06:35:00	06:35:00	stop_F	6
trip_1	06:40:00	06:40:00	stop_A	7

trips.txt

route_id	trip_id	service_id	block_id
red_loop	trip_1	weekday	red_loop_block

frequencies.txt

trip_id	start_time	end_time	exact_times	headway_secs
trip_1	6:10	18:40	1	1800

Notes

• My interpretation is that the approach above could also support interlining (Route 1 -> Route 2 -> Route 1, etc) if the trips don't overlap.
• But if headway_secs was less than the duration of trip_1 (30 min) then this would be an error. The spec doesn't support that right now.

[barbeau]

exact_times=1 trips defined in frequencies.txt should be treated the same way as trips defined in a GTFS that doesn't include the frequencies.txt file - you just "unroll" the pattern defined in stop_times.txt into individual trips from the start to end time defined in frequencies.txt, with the start time for each individual trip being headway_secs apart. Note that then arrival_time and departure_time in this case don't refer to absolute times, but rather exist to define the travel time between each stop in the trip. I agree that the documentation could be improved, including examples, to make this clearer.

Note there is another open proposal to better define in-seat transfers and transfer rules at https://github.com/google/transit/pull/32.

[derhuerst]

exact_times=1 trips defined in frequencies.txt should be treated the same way as trips defined in a GTFS that doesn't include the frequencies.txt file - you just "unroll" the pattern defined in stop_times.txt into individual trips [...].

Okay, thanks for clarification.

In this case, I advocate to state clearly in the documentation that one trip_id does not correspond to one "run" (which I tried to define above). From my subjective experience, this seems to be a quite natural assumption.

[antrim]

@derhuerst : I see what you mean. Perhaps a future modification to the spec or training materials could clarify this.

GTFS was created originally with passenger-facing applications in mind, so a "trip" refers to when a vehicle operates on a route. In passenger-facing information, that usually looks like a row on a timetable.

Operational schedules have runs, which would usually consist of multiple "trips" in the passenger-centric sense of GTFS.

Some (non-standard) GTFS datasets do include information on "runs" as you're thinking of them. Discussion in issue #195. Here is an example runcut.txt file: https://openmobilitydata.org/p/ventura-county-transportation-commission/792/latest/file/runcut.txt

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[skinkie]

Keep open.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[github-actions[bot]]

This issue has been closed due to inactivity. Issues can always be reopened after they have been closed.

[huntrob]

I'm currently looking at creating a frequencies.txt file for a rail operation. Having the main reference point as trip_id really threw me for a loop as that is not what I had expected to see there. After looking at this for a while, I determined that it uses a trip_id to pull the required data which I believe is route, stop sequence, and the running time. I think this should be clarified as it would save a lot of trial and error for other users.

[skinkie]

@huntrob consider this trip_id some kind of hash result using the same stop sequence, times between them and calendar. Then this template can be instantiated at different times.