Differences between revisions 5 and 19 (spanning 14 versions)
Revision 5 as of 2014-10-22 20:02:21
Size: 5845
Editor: JendrikSeipp
Comment: make script agnostic
Revision 19 as of 2023-11-22 09:03:02
Size: 6033
Editor: GabiRoeger
Comment:
Deletions are marked like this. Additions are marked like this.
Line 28: Line 28:
An exception are single-element lists, where the brackets can be dropped. For example, let `h1`, `h2`, `h3`
be heuristic specifications, then `[h1, h3]`, `[h2]` and `h2` are examples for a
For example, let `h1`, `h2`, `h3`
be heuristic specifications, then `[h1, h3]` and `[h2]` are examples for a
Line 45: Line 45:
 * Parameters of type `int` can by specified by "infinity". This means that the parameter will take the value numeric_limits<int>::max(), which is usually equal to 2^31 - 1.
Line 47: Line 46:
 * To get positions and keywords, run  * To get positions and keywords, use
Line 49: Line 48:
./downward-release --help [Name] --help [Name]
Line 52: Line 51:
== Integers ==
Parameters of type `int` can be specified as "infinity". This means that the parameter will take the value `numeric_limits<int>::max()`, which is usually equal to 2^31 - 1. If an `int` parameter value ends with "K", "M" or "G", the value is multiplied by one thousand, one million or one billion, respectively. For example, "bound=2K" is equivalent to "bound=2000".

== Strings ==
Parameters of type `string` can be specified in double quotes. Nested quotes can be escaped as `\"`, backslashes as `\\`, and newlines as `\n`.
E.g.,
{{{
filename="C:\\some.file"
}}}
Line 57: Line 65:
--heuristic "hff=ff()" --heuristic "hcea=cea()" \ --evaluator "hff=ff()" --evaluator "hcea=cea()" \
Line 61: Line 69:
instead of == Enumerations ==
Line 63: Line 71:
Enumeration arguments should be specified by name, e.g.:
Line 64: Line 73:
--heuristic "hff=ff()" --heuristic "hcea=cea()" \
--search "lazy_greedy(hff, hcea, preferred=(hff, hcea))" \
eager_greedy([h1,h2], cost_type=normal)
Line 68: Line 76:
Single element lists can be given without any brackets, e.g. To get enumeration names (and more), run
Line 70: Line 78:
--heuristic "hff=ff()" \
--search "lazy_greedy(hff, preferred=hff)" \
--help [Name] //e.g. with Name=eager_greedy
Line 74: Line 81:
== Enumerations ==
Enumeration arguments can now be specified by name or by number (previously only by number), e.g.
{{{
selmax([h1,h2], classifier=AODE)
}}}
and
{{{
selmax([h1,h2], classifier=1)
}}}
== Variables ==
Line 84: Line 83:
are equivalent. To get enumeration names (and more), run
{{{
--help [Name] //e.g. with Name=selmax
}}}
Often an object should be used for several purposes, e.g. a [[Doc/Evaluator|Heuristic]] or a [[Doc/LandmarkFactory|LandmarkFactory]]. The most
prevalent use case is a heuristic that is used for both the heuristic
estimates and for its preferred operators. In this case, one should define a variable for the object. We currently only support variables for
[[Doc/Evaluator|Heuristics]] and [[Doc/LandmarkFactory|LandmarkFactories]] but will extend the support for other feature types in the future.

Variables can be defined with {{{let(name, definition, expression)}}}
 * `name`: a variable name that should denote the feature
 * `definition`: an expression defining the value of the variable
 * `expression`: an expression defining any other feature. Occurrences of `name` in this expression may refer to the feature defined by `definition`.
Line 90: Line 94:
== Predefinitions ==

Often an object should be used for several purposes, e.g. a [[Doc/Heuristic|Heuristic]] or a [[Doc/LandmarkGraph|LandmarkGraph]]. The most
prevalent use case is a heuristic that is used for both the heuristic
estimates and for its preferred operators. In this case, one should predefine the object.

=== Heuristic Predefinitions ===

Heuristics can be predefined using the search option
`--heuristic` (see [[PlannerUsage#search]]).

{{{
--heuristic name=heuristic
}}}

 * `name` (string): a name that should denote the heuristic
 * `heuristic` ([[Doc/Heuristic|Heuristic]]): the heuristic

=== Landmark Predefinitions ===

If a set of landmarks should be used for several purposes, it can be predefined using the search option
`--landmarks` (see [[PlannerUsage#search]]) to avoid duplicate work and memory usage.

{{{
--landmarks name=landmarks
}}}

 * `name` (string): a name that should denote the set of landmarks
 * `landmarks` ([[Doc/LandmarkGraph|LandmarkGraph]]): the set of landmarks

=== Predefinition Example ===
=== Variable Example ===
Line 126: Line 100:
--landmarks "lm=lm_hm(m=2)"
--search "iterated([
    
lazy_greedy(lmcount(lm)),
    lazy_greedy(lmcount(lm,admissible=true))])"
--search "let(lm, lm_hm(m=2),
              iterated([lazy_greedy(lmcount(lm)),
                     lazy_greedy(lmcount(lm,admissible=true))]))"
Line 131: Line 104:


=== Old-style Predefinitions ===

We still support but deprecate the use of "predefinitions" before the {{{--search}}} option. They are internally converted to {{{let}}}-expressions.

The command lines
{{{
--evaluator name=definition --search expression
--landmarks name=definition --search expression
}}}
are both transformed to
{{{
--search let(name, definition, expression)
}}}
Line 139: Line 128:
--if-unit-cost --heuristic "h1=ff()" --heuristic "h2=blind()" \
--if-non-unit-cost --heuristic "h1=cea()" --heuristic "h2=lmcut()" \
--if-unit-cost --evaluator "h1=ff()" --evaluator "h2=blind()" \
--if-non-unit-cost --evaluator "h1=cea()" --evaluator "h2=lmcut()" \

Back to the HomePage.

Option Syntax

Meaning of the call syntax documentation

All parameters can be specified by keyword or by position. Once a parameter is specified by keyword, the rest of the parameters must be specified by keyword too. Some parameters have default values and are optional. These parameters are documented in the form keyword = defaultvalue.

Consider the following example:

name(p, qs, r, s=v1, t=Enum1)
  • p (type_p): some explanation

  • qs (list of type_q): some explanation

  • r (type_r): some explanation

  • s (type_s): some explanation

  • t (Enum): some explanation

    • Enum0: some explanation
    • Enum1: some explanation
    • Enum2: some explanation

Parameters p, qs and r are mandatory. qs is a list parameter. List parameters have to be enclosed in square brackets. For example, let h1, h2, h3 be heuristic specifications, then [h1, h3] and [h2] are examples for a list of heuristic specifications.

Parameters s and t are optional. s has the default value v1 and t the default value Enum1. t is an enumeration parameter and can only take the values listed (here Enum0, Enum1, Enum2). These values may also be passed by number, e.g. here t=Enum1 and t=1 are equivalent.

Some possible calls for this specification (with X and Xi having type_x):

  • name(P, Q, R): s and v have their default values v1 and Enum1

  • name(P, [Q], R): equivalent to previous call

  • name(P, [Q1, Q2], R, t=Enum2): s has its default value v1

  • name(t=1, r=R, qs=[Q1, Q2], s=S1, p=P) is equivalent to name(P, [Q1, Q2], R, S1, 1)

Notes

  • Parameters of type bool are specified by strings true or false

  • not case-sensitive
  • To get positions and keywords, use

--help [Name]

Integers

Parameters of type int can be specified as "infinity". This means that the parameter will take the value numeric_limits<int>::max(), which is usually equal to 2^31 - 1. If an int parameter value ends with "K", "M" or "G", the value is multiplied by one thousand, one million or one billion, respectively. For example, "bound=2K" is equivalent to "bound=2000".

Strings

Parameters of type string can be specified in double quotes. Nested quotes can be escaped as \", backslashes as \\, and newlines as \n. E.g.,

filename="C:\\some.file"

Lists

List arguments have to be enclosed in square brackets now. E.g.,

--evaluator "hff=ff()" --evaluator "hcea=cea()" \
--search "lazy_greedy([hff, hcea], preferred=[hff, hcea])" \

Enumerations

Enumeration arguments should be specified by name, e.g.:

eager_greedy([h1,h2], cost_type=normal)

To get enumeration names (and more), run

--help [Name]   //e.g. with Name=eager_greedy

Variables

Often an object should be used for several purposes, e.g. a Heuristic or a LandmarkFactory. The most prevalent use case is a heuristic that is used for both the heuristic estimates and for its preferred operators. In this case, one should define a variable for the object. We currently only support variables for Heuristics and LandmarkFactories but will extend the support for other feature types in the future.

Variables can be defined with let(name, definition, expression)

  • name: a variable name that should denote the feature

  • definition: an expression defining the value of the variable

  • expression: an expression defining any other feature. Occurrences of name in this expression may refer to the feature defined by definition.

Variable Example

Suppose I want to run GBFS with the lm_count heuristic (the inadmissible version), and then run another GBFS search with an admissible lm_count heuristic, using the h^m landmarks without discovering the landmarks twice.

--search "let(lm, lm_hm(m=2), 
              iterated([lazy_greedy(lmcount(lm)),
                        lazy_greedy(lmcount(lm,admissible=true))]))" 

Old-style Predefinitions

We still support but deprecate the use of "predefinitions" before the --search option. They are internally converted to let-expressions.

The command lines

--evaluator name=definition --search expression
--landmarks name=definition --search expression

are both transformed to

--search let(name, definition, expression)

Conditional options

In some cases, it is useful to specify different options depending on properties of the input file. For example, the LAMA 2011 configuration makes use of this, adding an additional cost-ignoring search run at the start for tasks with non-unit action costs.

Example

--if-unit-cost --evaluator "h1=ff()" --evaluator "h2=blind()" \
--if-non-unit-cost --evaluator "h1=cea()" --evaluator "h2=lmcut()" \
--always --search "eager_greedy([h1, h2])"

This conducts an eager greedy search with two heuristics. On unit-cost tasks, it uses the FF heuristic and the blind heuristic. On other tasks, it uses the context-enhanced additive heuristic and the LM-Cut heuristic.

Details

Options can be made conditional via selectors such as --if-unit-cost. All options following a selector are only used if the condition associated with the selector is true. (This really includes all options, including ones like --plan-file that do not affect the planning algorithm.) Each selector is in effect until it is overridden by a new selector. The following selectors are available:

  • --if-unit-cost: the following options are only used for unit-cost planning tasks (i.e., tasks where all actions have cost 1, including the case where no action costs are specified at all)

  • --if-non-unit-cost: opposite of --if-unit-cost

  • --always: the following options are always used

FastDownward: OptionSyntax (last edited 2023-11-22 09:03:02 by GabiRoeger)