search

opengeospatial / sensorthings

The official web site of the OGC SensorThings API standard specification.
134 stars 29 forks source link

Temporal Logic #2

Open hylkevds opened 8 years ago

hylkevds commented 8 years ago

For one of our projects we're using predictions, where each observation has a validTime interval. I just noticed the SensorThings API doesn't define any functions to query time intervals.

The most common query here would be "give me all observations that are valid at this time point". Something like v1.0/Observations?$filter=during(2016-07-26T09:56:00Z, validTime) or v1.0/Observations?$filter=contains(validTime, 2016-07-26T09:56:00Z) But the best I can think of with the current set is v1.0/Observations?$filter=not validTime gt 2016-07-26T09:56:00Z and not validTime lt 2016-07-26T09:56:00Z

But there are several more functions that would be very useful:

hylkevds commented 7 years ago

I've updated the suggested changes a bit: https://github.com/hylkevds/SensorThingsAPI/compare/3e5bcc9c662ae24e6054d4823406c3ba83de35ae...master and also created tests for the conformance test suite: https://github.com/hylkevds/ets-sta10/blob/libraryTests/src/main/java/org/opengis/cite/sta10/filteringExtension/DateTimeTests.java

hylkevds commented 7 years ago

I've grouped the changes for this issue in a single branch: https://github.com/hylkevds/SensorThingsAPI/commits/timeFunctions

hylkevds commented 7 years ago

A diff of the change is here: https://github.com/hylkevds/SensorThingsAPI/compare/reworkRoot...timeFunctions

liangsteve commented 7 years ago

@hylkevds let's use the same operations and definitions of OGC Filter Encoding's temporal operator. OGC 09-026r1 Filtering Encoding.

hylkevds commented 7 years ago

Hmm, OGC 09-026r1 Filtering Encoding points to iso 19108 temporal operators, which is not free. Assuming GeoTools implements the filters correctly, there are a few differences:

OGC 09-026R1 iso 19108 STA proposal comment geotools link
After After after After
Before Before before Before
Begins Begins starts Proposal is wider Begins
BegunBy Begun -
TContains Contains -
During During during iso: BothOpen. Proposal: half-open During
Ends Ends finishes proposal is wider Ends
TEquals Equals equals TEquals
TOverlaps Overlaps overlaps proposal: also for instants TOverlaps
Meets Meets meets iso: 1<2. Sta: both ways, also instants Meets
OverlappedBy OverlappedBy -
MetBy MetBy -
EndedBy EndedBy -
AnyInteracts N/A - same as overlaps in proposal AnyInteracts

First of all, the proposal leaves out the inverse operations. I find they just clutter the standard.

Begins / ends: For two intervals, iso 19108 requires the first interval to be shorter than the second. In our proposal, the length of the intervals doesn't matter, as long as they start/end at the same time. The proposal uses the names "starts/finishes" but that is a small change.

During: iso 19108 uses both-open intervals. This means that if we take a series of contiguous intervals (t1-t2, t2-t3, t3-t4, t4-t5) then iso 19108 defines t1, t2, t3, t4 and t5 not to be "during" any of these intervals. I think that is highly unexpected for users. Therefore the proposal defines intervals to be half-open, with the starting time being part of the interval, but the end-time not.

Overlaps: iso 19108 defines overlaps only for intervals. The proposal also allows instants.

Meets: iso 19108 defines that the first of the two intervals has to come first (t1.end = t2.start). The proposal also allows the other way around.

What do you all think?

liangsteve commented 6 years ago

Note from STA SWG at Southhapton TC: ask Chris Little and/or Simon Cox regarding ISO time standards. Especially on why overlaps is both open. Seems counter-intuitive.

Will vote in SWG after asking the experts.

hylkevds commented 6 years ago

I think ISO defines times to always be intervals, meaning that a time "instant" actually has a duration, and the given time instant is actually the start of a very short interval.

If that's so, then the result is that intervals are half-open, with the start being part of the interval. A time instant at a boundary between two contiguous intervals will overlap with the second interval.

In that case, an info box explaining the half-open behavior would be sufficient.

hylkevds commented 5 years ago

In the pull request I've added the time functions as defined in OGC 09-026r1 Filtering Encoding