Skip to content

parameters.md

Latest commit

 

History

History
131 lines (103 loc) · 11.1 KB

parameters.md

File metadata and controls

131 lines (103 loc) · 11.1 KB
Raw

Parameter Reference

Parameters are read from a JSON file (e.g. AnyParameters.json or BatchParameters.json) and a scenario is selected on the command line:

--paramfile <name> --scenario <key>

Each scenario block sets only the keys it needs to override; every other parameter falls back to the code default listed below. The defaults and JSON key names in this document are taken directly from ReadWrite::readParametersJson (in ReadWrite.cpp), and the option values correspond to the enums in Types.h. Where a paper uses a different value for an experiment, it is noted in the Notes column.

Conventions

  • LARGE_CONSTANT = 1e7 is used as a sentinel for "∞ / disabled" (see Types.h).
  • Booleans are written in JSON as 1 (true) / 0 (false).
  • Symbols in the Notes column refer to the notation used in the two papers (B–CG: Accelerated Column Generation; A–CG: Anytime Optimization Approach for Dynamic DARP).

Objective Weights

Key Type Default Description Notes
Wait_W1 float 1.0 Weight on total passenger waiting time in the objective ωwait
Ride_W2 float 0.0 Weight on trip delay (excess ride time = ride time − direct travel time) ωdrop; papers use 0.5 (best trade-off) and 1.0
Req_W3 bool 0 Weight each objective term by the number of passengers q_i of the request (1 = enabled) Appendix G of B–CG (per-passenger weighting)
Ride_W4 bool 0 Use total in-vehicle ride time instead of excess ride time (1 = enabled) Jung et al. variant (λexcess=0)
Relative_W5 bool 0 Normalize the ride-time term by the direct travel time t_i → relative detour (1 = enabled) Pfeiffer & Schulz variant (λnormal=t_i)
Normal_W6 bool 0 Express/average the objective per served customer (1 = enabled) Reported in code as "objective per customer"

The default objective (Wait_W1=1, Ride_W2=0) reproduces the waiting-time-only objective of Riley et al. (2019). Setting Ride_W2=0.5 adds the trip-delay term used throughout both papers.

Model Parameters and Algorithmic Setting

Key Type Default Description Notes
alphaParam float 1.5 Detour factor α: contributes to max ride time t_max = max(α·t_i, β + t_i) α
betaParam float 240.0 Additive ride-time slack β (seconds) in t_max = max(α·t_i, β + t_i) β; Set 2 / shuttle experiments use 300–600
deltaPram float 420.0 Base penalty (in equivalent seconds) for an unserved request: p_i = δ · 2^((τℓ − e_i)/(penaltyL·ℓ)) ρ in the papers; shuttle scenarios use 600
penaltyL int 30 Epoch-length term used in the unserved-request penalty growth formula denominator scaling in p_i
epochLength float 30.0 Length of a fixed epoch / time slot (seconds); used in DYNAMIC mode (B–CG) ℓ
committedTime float 30.0 Length of the commitment horizon (seconds); used in ANYTIME mode (A–CG) Lcmt; starts at 30 s, grows in 5 s steps
NumIter int 1 Max number of CG iterations per epoch (passive stopping policy η) A–CG uses η = 1
informTimeLimit float ∞ (1e7) Waiting time after which the passenger is notified of a committed pickup time (seconds) Tnotif; experiments use 60
pickupDeviationWindow float ∞ (1e7) Allowable deviation around a committed pickup time (seconds) ε; experiments use 60
nbThreads int 16 Number of threads for parallel subproblem solving

Vehicle Rebalancing / Fleet

Key Type Default Description Notes
vehicleReturn bool 0 Enable idle-vehicle repositioning toward high-demand areas (1 = enabled) auxiliary rebalancing
MaxWait float ∞ (1e7) Waiting-time threshold θ that marks a committed request as "late" for repositioning (seconds) θ; experiments use 120–360, calibrated to 300 (5 min)
WaitForReturn float ∞ (1e7) Idle inactivity period before a vehicle becomes eligible for repositioning (seconds) experiments use 240
returnType int 1 Relocation policy: 0 = TO_SOURCE (return to origin depot), 1 = ASSIGN (return based on assignment)

Column Generation

Key Type Default Description Notes
NumColumn int 50 Max number of columns (routes) added to the RMP per vehicle per CG iteration Ω
sortColumn int 1 Column ordering before adding to RMP: 0 = NORMAL_RC, 1 = RC (reduced cost), 2 = LAMBDA_S, 3 = COMP_C
Route_Recycle bool 0 Reuse routes from the previous epoch as a warm start / persistent pool (1 = enabled) "keep" scenarios
reducedCostThreshold double 100.0 Reduced-cost threshold γ above which pooled columns are pruned (used in BY_POOL reuse)
InitialDual int 1 Dual initialization: 0 = PENALTIES, 1 = LAST_LP, 2 = BARRIER, 3 = INITIAL_LP, 4 = GREEDY_D, 0 = cold start, 1 = dual warm start
DualMethod int 0 Dual extraction method: 0 = LMP, 1 = INTERIOR
SmoothDual bool 0 Apply smoothing to dual variables between CG iterations (1 = enabled)
warmStart int 1 Primal warm-start strategy: 0 = GREEDY_START, 1 = PRE_SOLUTION, 2 = EMPTY_ROUTES
MIPGap float 0.001 MIP relative optimality gap (0.001 = 0.1 %)
solutionMode int 1 Solve mode: 0 = STATIC (offline), 1 = DYNAMIC (B–CG / fixed-epoch batching), 2 = ANYTIME (A–CG / flexible horizon)
mainAlgorithm int 0 Algorithm variant: 0 = GREEDY, 1 = MIP, 2 = B_CG, 3 = F_ICG, 4 = A_CG

ISUD / MIP Solver

Key Type Default Description Notes
isudVariant int 1 ISUD solve strategy: 0 = ISUD_MIP_RP, 1 = ISUD_PIVOT_RP
reducedCP bool 0 Use a reduced constraint set in the CP step (1 = enabled)
minImp float 0.0025 Minimum improvement threshold for ISUD
BigM int 27000 Big-M value used in MIP formulations key spelled BigM
solveTimeLimit int 300 Time limit for solving MIP models (seconds)
populateTimeLimit int 200 Time limit for populating MIP models (seconds)

Subproblem / Labeling

Key Type Default Description Notes
subproblemAlgorithm int 1 Subproblem solver: 0 = MIP_SUB, 1 = LABEL_SETTING
nbPick int 4 Maximum number of pickups per route in labeling MPick; A–CG fixes 2, B–CG uses dynamic up to 4
Dynamic_Pricing bool 0 Gradually increase the pickup limit nbPick across SP iterations (1 = enabled) mutually exclusive with Partial_Pricing
Partial_Pricing bool 0 Solve the subproblem for a subset of vehicles per iteration (1 = enabled) mutually exclusive with Dynamic_Pricing
pruneNodes bool 1 Remove nodes that cannot improve the best solution (1 = enabled) Corollary 1
pruneArcs bool 1 Remove arcs that cannot improve the best solution (1 = enabled) Corollary 2
discardSuboptimalPath bool 1 Discard partial paths dominated / outside the penalty window (1 = enabled) Corollary 3
isDropPickPossible bool 0 Allow visiting a pickup after a drop-off on the same route (1 = enabled) "drop–pick" restriction is the inverse
isTruncated bool 1 Truncate the label search when MaxLabel is reached (1 = enabled)
MaxLabel int 15 Max number of labels retained per node during truncated labeling MLabel
MaxCommittedLabel int 0 Max number of committed labels retained per node during truncated labeling
sortPath int 0 Label ordering for truncation: 0 = L_SCORE (normalized reduced cost), 1 = RD_COST (reduced cost), 2 = LAMBDA (lambda score) 0 recommended (B–CG)
LabelingStrategy int 1 Node traversal strategy: 0 = PUSHING, 1 = PULLING
isDominanceReleased bool 0 Relax the last condition of the dominance rule to dominate more labels (1 = enabled) "relaxed dominance"
reoptimizeSP bool 0 Re-optimize subproblems or solve from scratch (1 = enabled)
LabelingReOptimizeStrategy int 2 Reduced-network / reuse mode: 0 = RE_INSERT, 1 = BY_BASIS, 2 = BY_POOL A–CG compares BY_BASIS vs BY_POOL
newRequestLimit int ∞ (1e7) Threshold on the number of new requests that triggers re-optimization

Validation Rules

The parser rejects invalid combinations:

  • Dynamic_Pricing and Partial_Pricing cannot both be enabled in the same scenario.
  • A scenario name is required (--scenario <key>); the parser throws if it is missing or not found under the scenarios object.