Interval Estimation

[mority]

Initial

• performs an initial extension of the search interval based on the requested number of connections
• in case of forward search, the interval is extended by 1 hour into the past and by 1 hour/connection into the future
• in case of backward search, the interval is extended by 1 hour into the future and by 1 hour/connection into the past

Extension

• the interval is extended by its current size times the number of connections that are still required to meet the requested minimum

[felixguendling]

Ready to merge? Can you please copy&paste the benchmark results? (just for reference)

[mority]

I ran the following benchmarks on my PC at home on the DELFI timetable of Germany. I can rerun them on Turing to get some numbers for the transitous dataset if you like.

Baseline, 1 hour initial size

--- routing times [ms] --- (n = 1000)
  25%:          856
  50%:         1630
  75%:         3906
  90%:         8678
  99%:        34407
99.9%:       333043
  max:       333043
----------------------------------

--- search iterations --- (n = 1000)
  25%:            1
  50%:            2
  75%:            5
  90%:           13
  99%:         6060
99.9%:         9048
  max:         9048
----------------------------------

Baseline, 2 hours initial size

--- routing times [ms] --- (n = 1000)
  25%:          694
  50%:         1483
  75%:         3639
  90%:         8404
  99%:        34968
99.9%:       328875
  max:       328875
----------------------------------

--- search iterations --- (n = 1000)
  25%:            0
  50%:            1
  75%:            4
  90%:           12
  99%:         6059
99.9%:         9047
  max:         9047
----------------------------------

Interval Estimate

--- routing times [ms] --- (n = 1000)
  25%:          760
  50%:         1396
  75%:         2891
  90%:         5690
  99%:        24330
99.9%:       161848
  max:       161848
----------------------------------

--- search iterations --- (n = 1000)
  25%:            0
  50%:            0
  75%:            1
  90%:            2
  99%:            6
99.9%:            6
  max:            6
----------------------------------

[mority]

Benchmarks on Turing and the transitous dataset

Baseline, 1 hour initial size

--- routing times [ms] --- (n = 1000)
  25%:         4783
  50%:        17475
  75%:        97642
  90%:       664527
  99%:      2504066
99.9%:      3515880
  max:      3515880
----------------------------------

--- search iterations --- (n = 1000)
  25%:            2
  50%:           10
  75%:         6186
  90%:         7215
  99%:         8587
99.9%:         8727
  max:         8727
----------------------------------

Baseline, 2 hour initial size

--- routing times [ms] --- (n = 1000)
  25%:         4430
  50%:        17613
  75%:        97525
  90%:       652633
  99%:      2486320
99.9%:      3667003
  max:      3667003
----------------------------------

--- search iterations --- (n = 1000)
  25%:            2
  50%:           10
  75%:         6185
  90%:         7214
  99%:         8586
99.9%:         8726
  max:         8726
----------------------------------

Interval Estimate

--- routing times [ms] --- (n = 1000)
  25%:         2828
  50%:         8862
  75%:        52513
  90%:       292759
  99%:      1379750
99.9%:      2810408
  max:      2810408
----------------------------------

--- search iterations --- (n = 1000)
  25%:            0
  50%:            2
  75%:            6
  90%:            6
  99%:            6
99.9%:            8
  max:            8
----------------------------------