lower bounds greedy RAPTOR

[mority]

Building a RAPTOR that uses lower bounds as goal direction

• At the beginning of each round when iterating the station marks to assign route marks. Each route is assigned the lower bounds value of that station. This is stored in vector_map<route_idx_t, std::uint16_t> route_lb_, same thing for RT transports.
• When handling the route marks, we retrieve the lower bounds value for each stop. If the lower bounds value of the stop is bigger than the one stored in route_lb_ the stop is skipped (again we do the same thing for RT transports).
• Enabled it for the ping of PONG but it leads to a lot failures where the pong can not find the journey of the ping. Therefore, I am only working with search.h, i.e. range RAPTOR, for now.

versions so far:

• greedy: skip stops with greater lower bounds value
• more greedy: skip stops with greater or equal lower bounds value
• acceptable loss factor: skip stops with a lower bounds value that is greater by some factor
• threshold: do not skip if lower bound value is under a certain threshold

Unit Tests

The implementation passes most unit tests for the vanilla RAPTOR. Current failures:

• td_offsets
• some via tests

Evaluation on DELFI GTFS

Correctness

with interval extension:

• greedy: 326/1000
• more greedy: 35/1000

fixed 2h search windows:

• acceptable loss 1.0: 478/1000
• acceptable loss 1.42: 738/1000
• more greedy: 198/1000
• acceptable loss 1.42, threshold 60: 966/1000

Performance: with array extension and min connection count = 5

baseline

response_time
average: 1,381.23
max: 37,397
99 quantile: 19,717
90 quantile: 3,227
80 quantile: 1,685
50 quantile: 352
min: 0

greedy

response_time
average: 663.49
max: 27,799
99 quantile: 5,231
90 quantile: 1,414
80 quantile: 796
50 quantile: 312
min: 1

more greedy

response_time
average: 691.30
max: 9,232
99 quantile: 4,581
90 quantile: 1,518
80 quantile: 964
50 quantile: 359
min: 2

Performance: 2h search windows and no min con count

baseline

response_time
average: 394.78
max: 4,868
99 quantile: 2,549
90 quantile: 1,080
80 quantile: 615
50 quantile: 155
min: 0

acceptable loss = 1.0 (i.e., no increase of lb accepted)

response_time
average: 228.58
max: 2,601
99 quantile: 1,195
90 quantile: 452
80 quantile: 292
50 quantile: 145
min: 0

acceptable loss = 1.42 (i.e., the lb may increase up until 42%)

response_time
average: 314.44
max: 3,797
99 quantile: 2,120
90 quantile: 839
80 quantile: 413
50 quantile: 135
min: 0

more greedy

response_time
average: 136.27
max: 703
99 quantile: 400
90 quantile: 223
80 quantile: 169
50 quantile: 109
min: 0

acceptable loss 1.42, threshold 60

response_time
average: 324.71
max: 3,858
99 quantile: 2,164
90 quantile: 848
80 quantile: 435
50 quantile: 146
min: 0