API.Planning.Domains

An interface to the repository of PDDL domains and problems.

Description

The planning.domains API provides programmatic access to a wide collection of PDDL benchmark domain and problem files. All of the physical files can be found in the repository of domains over at GitHub. You can get a copy for yourself by cloning the repository, or using the command-line utility described below.

If you find any bugs in the PDDL files (incomplete sets, bad formulations, etc), clone the repository and submit your fix as a pull request. Where it makes sense, we will also accept pull requests with entirely new benchmark sets (e.g., after new IPC contests or as a release of a particular suite). For now, only classical PDDL benchmarks will be accepted (of any PDDL level / expressivity), but in the future we hope to expand this to other richer formalisms (POND, FOND, RDDL, etc).

The collection of files is just the first component of api.planning.domains. The primary aspect of this service is the interface offered to query, view, and explore the existing domains. We store, and continually update, attributes for each of the domains and problems. For example, the lower and upper bounds on plan cost, the requirements found in a domains, the classical width of a problem, etc. On this page, you can find a description of the interface to api.planning.domains, and details on the tools / libraries constructed to make interacting with the API even easier.

The API contains three types of objects:

  1. Problems: Information for each individual problem, including its file, the corresponding domain file, statistics on the problem, etc.
  2. Domains: Information for each individual domain, including its description and various statistics for the domain.
  3. Collections: Sets of domains, including a collection for every IPC, individual planner collections, etc.

Disclaimer: The API should be considered to be in a very alpha phase at the moment. The hope is to eventually have a way for citing precisely the benchmarks used in evaluating your software, but for the time being we are still fixing up the domains, replacing problem sets, and correcting bugs in the various statistics. Until this disclaimer is removed, we do not recommend using the API for official business (academic or otherwise).

Thanks: The initial statistics in the database for the API were due to the efforts of Miquel Ramirez, Patrik Haslum, and Nir Lipovetzky. We are working on a wiki-like interface for anyone in the community to contribute if they wish. If you would like to see a new type of statistic, or would like to help automatically generate existing ones, let me know or create an issue on our bug tracker.

API Interface

API URL (back)


API Tools

Purpose

To demonstrate what is possible with the API, and to provide a more practical interface to the data, we have created a set of tools which can be found in the api-tools repository.

Python Library

The planning_domains_api.py library provides access to the API through the following functions (more to be added in the future):

Method Description
query(qs)Query for the string http://api.planning.domains/json/classical/{qs}
simple_query(qs)Helper function that calls query(qs) and wraps it with a bit of error handling. This is recommended over query(qs).
get_collections()Return the list of collections.
get_collection(cid)Get the collection object corresponding to the collection id cid.
find_collections(name)Return the list of collections with a name matching the passed in argument.
get_domains(cid)Get the list of domain objects that are in collection cid.
get_domain(did)Get the domain object corresponding to the domain id did.
find_domains(name)Return the list of domains with a name matching the passed in argument.
get_problems(did)Get the list of problem objects that are in domain did.
get_problem(pid)Get the problem object corresponding to the problem id pid.
find_problems(name)Return the list of problems with a name matching the passed in argument.
localize(prob)Changes the file paths in a problem object to the appropriate local path assuming you have set it using the planning.domains.py utility. Note that problems are localized by default when using the api (e.g., when using get_problems).
get_tags()Get the list of all tags.
tag_TYPE(id,tname)Update a TYPE(collection, domain, or problem) with a given id and tagname.
untag_TYPE(id,tname)Remove the tag from a TYPE(collection, domain, or problem) with a given id and tagname.
update_TYPE_stat(...)Updates an API attribute for a given TYPE(collection, domain, or problem) given the following parameters: (iid) item id; (attribute) name of the attribute; (value) new value; (description) info about the update. An additional helper function exists to find all problems with a missing attribute: get_null_attribute_problems(attribute)
get_plan(pid)Get the IPC-style plan for problem pid.
submit_plan(pid,plan)Submit a new IPC-style plan for problem pid.

Example that demonstrates how to test on every problem in the "All IPC" set (note, this assumes that you have already run the command line utility to set up the location of the planning. domains repository):

import sys

print "Loading domains...",
sys.stdout.flush()

import planning_domains_api as api

# 12 is the collection for all STRIPS IPC domains
domains = {}
for dom in api.get_domains(12):

    # Turn the links into relative paths for this machine
    probs = api.get_problems(dom['domain_id'])

    # Map the domain name to the list of domain-problem pairs
    domains[dom['domain_name']] = []
    for p in probs:
        domains[dom['domain_name']].append((p['domain_path'], p['problem_path']))

print "done!"

Command-line Utility

The planning.domains.py command-line utility allows you to setup your repository of domains and do primitive queries. It will check out the domains repository if you wish, and store a local settings file so that it can be automatically retrieved using the planning_domains_api.py library.

The local settings file is also used by the API in order to store the user's email and (if available) access token for making changes to the API attributes.

The tool is interactive to set things up, so just run ./planning.domains.py from the command line to get started. You can also run it without parameters to get the usage:

> ./planning.domains.py

 No command-line options given.  Usage:

 planning.domains.py update                                Update the local domain repository.

 planning.domains.py register                              Register your email and token for making API edits

 planning.domains.py find collections [string]             Find collections whose title/ID contains 'string'
 planning.domains.py find domains [string]                 Find domains whose title/ID contains 'string'
 planning.domains.py find problems [string]                Find problems whose title/ID contains 'string'

 planning.domains.py show collection [integer]             Find collections whose title/ID contains 'integer'
 planning.domains.py show domain [integer]                 Find domains whose title/ID contains 'integer'
 planning.domains.py show problem [integer]                Find problems whose title/ID contains 'integer'
 planning.domains.py show plan [integer]                   Show the plan (if any) matching the given problem ID

 planning.domains.py list collections                      Lists all of the collections.
 planning.domains.py list tags                             Lists all of the available tags.
 planning.domains.py list null-attribute [string]          Lists all of the problems that have a null attribute setting (string)

 planning.domains.py tag collection [integer] [string]     Tags the specified collection (integer) with a tag (string)
 planning.domains.py tag domain [integer] [string]         Tags the specified domain (integer) with a tag (string)
 planning.domains.py tag problem [integer] [string]        Tags the specified problem (integer) with a tag (string)
 planning.domains.py untag collection [integer] [string]   Un-tags the specified collection (integer) with a tag (string)
 planning.domains.py untag domain [integer] [string]       Un-tags the specified domain (integer) with a tag (string)
 planning.domains.py untag problem [integer] [string]      Un-tags the specified problem (integer) with a tag (string)

 planning.domains.py submit plan [integer] [plan file]     Submit the provided plan for validation and possible storage

 planning.domains.py generate-lab [integer] [string]       Generate a lab-style python file (string) given a collection (integer)
                    

Javascript Library

The planning-domains.js library makes it quick and easy to display problems, domains, and collections on a website. Both jQuery and Bootstrap are prerequisites, but other than that all you need to do is include the JavaScript file somewhere on your page:

<script type="text/javascript" src="planning-domains.js"></script>

Once loaded, you can call any of the following JavaScript functions to populate a div with the appropriate HTML:

  • insert_navigator(parent_div, selection_func_name)

    Inserts a full navigator of the collections / domains / problems in the API as the HTML text of parent_div. The function matching the name selection_func_name will be invoked with a single parameter corresponding to the problem ID once the user has navigated to a problem and selected it. The selection_func_name parameter is optional, and to see it in action, use the "Import" functionality of editor.planning.domains.

  • fetch_collections(query_string, parent_div, selection_func_name)

    Populates parent_div with a table of collections matching the query string query_string (e.g., '/collections'). The option selection_func_name will be used as a callback if a row corresponding to a collection is selected, and the collection ID will be used as a parameter.

  • fetch_domains(query_string, parent_div, selection_func_name)

    Populates parent_div with a table of domains matching the query string query_string (e.g., '/domains/12'). The option selection_func_name will be used as a callback if a row corresponding to a domain is selected, and the domain ID will be used as a parameter.

  • fetch_problems(query_string, parent_div, selection_func_name)

    Populates parent_div with a table of problems matching the query string query_string (e.g., '/problems/search?domain_name=blocks'). The option selection_func_name will be used as a callback if a row corresponding to a problem is selected, and the problem ID will be used as a parameter.

For examples, see the /web directory of the API Tools Repository.