Parameters

We provide a configuration file for PathWyse in the root directory: pathwyse.set.

If the file is present in the folder where PathWyse is run, it will override the library default configuration. Note that this is the current working directory of the process, not the folder containing the executable: running a binary built under a build/ folder from within that folder will fall back to the defaults unless pathwyse.set is also placed there (or the binary is run from the folder that contains it), showing a warning.

Settings are specified with the following format (case sensitive):

parameter = value

Values can be either numbers or text. Single-line comments are supported and must begin with #.

Solver

Parameter	Description	Values	Default
`verbosity`	Controls console output	`0` off · `1` standard · `2` adds solver operations · `3` data collection · `4` debug	`2`

Problem

Parameter	Description	Values	Default
`problem/memory_threshold`	Enables memory compression and sparse data structures above this node count	positive integer	`10000`
`problem/decimal_digits`	Decimal places to preserve when scaling floating-point costs to integers	non-negative integer	`2`
`problem/coordinates/distance/scaling`	Multiplier applied to computed distances	double	`0.95`
`problem/coordinates/microdegrees`	Scales coordinates from microdegrees to degrees	`0` no · `1` yes	`0`

Preprocessing

Parameter	Description	Values	Default
`algo/preprocessing/intensity`	Number of Dijkstra-based preprocessing rounds	`off` · `standard` (4 rounds) · `full` (standard + one round per extra resource)	`standard`

Algorithms

Parameter	Description	Values	Default
`algo_names`	Comma-separated list of algorithms to use	see below	`PWDefault`
`algo_active`	Comma-separated enable flags, one per algorithm in `algo_names`	`0` disabled · `1` enabled	`1`
`algo_parallel`	Runs all enabled algorithms in parallel	`0` sequential · `1` parallel	`0`

Exact algorithms

Name	Description
`PWDefault`	Multi-resource DP algorithm for cyclic and acyclic networks
`PWBucket`	Multi-resource bucket-based DP algorithm for cyclic networks
`PWAcyclic`	Single-resource A*-based DP algorithm for acyclic networks

Heuristic algorithms

Name	Description
`PWDefaultRelaxDom`	PWDefault with relaxed dominance
`PWDefaultRelaxQueue`	PWDefault with fixed-size label priority queues
`PWBucketRelaxDom`	PWBucket with relaxed dominance

Example: declare three algorithms, with only the two heuristics active at startup:

algo_names   = PWDefault,PWDefaultRelaxDom,PWDefaultRelaxQueue
algo_active  = 0,1,1
algo_parallel = 1

Algorithm IDs are assigned in order: PWDefault gets id 0, PWDefaultRelaxDom gets id 1, PWDefaultRelaxQueue gets id 2. These IDs are used at runtime to enable, disable, or reset individual algorithms via the Python interface.

PWDefault

Parameter	Description	Values	Default
`algo/default/autoconfig`	Autoconfigures best settings for the problem type	`0` off · `1` on	`1`
`algo/default/timelimit`	Time limit in seconds (`0.0` = no limit)	double ≥ 0	`0.0`
`algo/default/search`	Search direction	`bidirectional` · `forward` · `backward`	`bidirectional`
`algo/default/parallel`	Runs forward and backward steps in parallel (requires `bidirectional` search)	`0` off · `1` on	`0`
`algo/default/bidirectional/split`	Forward budget fraction of critical resource UB (requires `bidirectional` search)	`[0.0, 1.0]`	`0.5`
`algo/default/reserve`	Maximum labels stored per direction; memory is reserved upfront	positive integer	`10000000`
`algo/default/use_visited`	Labels track visited nodes (overridden by autoconfig)	`0` off · `1` on	`1`
`algo/default/compare_unreachables`	Include unreachable nodes in dominance checks (requires `use_visited = 1`, overridden by autoconfig)	`0` no · `1` yes	`1`
`algo/two_cycle_elimination`	Enables two-cycle elimination	`0` off · `1` on	`0`
`algo/default/ng`	NG-path relaxation type (does not guarantee an elementary solution)	`off` · `restricted` · `standard`	`off`
`algo/default/ng/set_size`	NG neighbourhood size per node (requires NG enabled)	positive integer	`8`
`algo/default/dssr`	DSSR type (guarantees optimality if completed before timelimit; applied after NG if both are enabled)	`off` · `restricted` · `standard`	`standard`
`algo/default/candidate/type`	Label candidate selection strategy (overridden by autoconfig)	`node` · `roundrobin`	`node`
`algo/default/join/type`	Join strategy after bidirectional search (requires `bidirectional` search)	see below	`ordered`
`algo/default/relaxations/queue_limit`	Node pool size limit for `PWDefaultRelaxQueue`	positive integer	`20`
`algo/requested_solutions`	Maximum solutions returned (PWDefault, PWBucket and relaxations)	positive integer	`1`

Example: setting algo/requested_solutions = 5 requests up to 5 solutions. The behaviour depends on the search configuration:

Bidirectional search: the k-ordered join is used, collecting the k best complete paths found during the join phase.
Unidirectional search (forward or backward): the k best labels reaching the destination are returned.

When relaxations are active, the returned solutions are the best k feasible paths found within the relaxation:

NG relaxation: the k best NG-routes are returned. Since NG-routes are a relaxation of the elementary constraint, some returned paths may contain repeated nodes and are not guaranteed to be elementary.
DSSR: the k best elementary solutions found up to that DSSR iteration are returned.

In both cases, some candidate solutions may be discarded before being returned: during the collection phase, paths that violate the relaxation constraints (e.g., contain forbidden cycles when using NG, or are non-elementary when using DSSR) are filtered out. As a result, fewer than k solutions may be returned.

Join methods

Value	Description
`naive`	Join all label pairs on arcs
`naive_node`	Naive join performed on nodes
`classic`	Use incumbent to prune suboptimal arc comparisons
`classic_node`	Classic join performed on nodes
`ordered`	Sort pools and terminate early when no improvement is possible (arcs)
`ordered_node`	Ordered join performed on nodes
`kordered`	Ordered join extended to collect multiple solutions (arcs); automatically selected when `requested_solutions > 1`
`kordered_node`	k-ordered join performed on nodes
`pareto`	Binary-search based join on arcs
`pareto_node`	Binary-search based join on nodes

PWBucket

PWBucket also reads the following parameters from the PWDefault section: algo/default/timelimit, algo/default/search, algo/default/parallel, algo/default/ng, algo/default/ng/set_size, algo/default/dssr, algo/two_cycle_elimination, algo/requested_solutions.

Parameter	Description	Values	Default
`algo/bucket/resource`	Resource ID used to index the buckets	non-negative integer	`0`
`algo/bucket/size_mode`	Bucket size strategy	`unit_size` (size 1) · `min_res_consumption` (minimum res consumption)	`unit_size`
`algo/bucket/prev_dom_check_percentage`	Number of preceding buckets checked for dominance, as a fraction of the total bucket count	`[0.0, 1.0]`	`1.0`
`algo/bucket/next_dom_check_percentage`	Number of following buckets checked for dominance, as a fraction of the total bucket count	`[0.0, 1.0]`	`1.0`

Data Collection and Output

Parameter	Description	Values	Default
`data_collection/level`	Extent of data collected	`-1` off · `0` performance data, no overhead · `1` adds insertion times, small overhead · `4` adds label collection (large overhead, requires `output/write = 1`)	`-1`
`output/write`	Write collected data to disk (in the output folder)	`0` off · `1` on	`0`