API-Specification¶

prac comes with an easy-to-use API, which lets you use the modules provided by prac of which each solves an own inference task conveniently in your own applications.

PRACModules¶

A PRACModule is represented by an instance of the class prac.core.base.PRACModule. Each module provides a predicates.mln file containing the predicate declarations for the MLN of the respective module and a pracmodule.yaml file specifying module properties such as its name and a description, as well as dependencies to other PRACModules, the names of the module’s class and a file (usually predicates.mln) containing the predicate declarations for the related MLN. It also specifies which project file contains the default project to be loaded for this module. This file will be processed during initialization of the PRAC reasoning system.

PRAC¶

The PRAC reasoning system is represented by an instance of the class prac.core.base.PRAC. This class encapsulates information about the modules of the PRAC pipeline and the ontology used for the inference. It also provides a method run to execute a module of the PRAC pipeline.

PRACInference¶

An instance of the class prac.core.inference.PRACInference represents an inference chain in PRAC which is initialized with a reference to an instance of the PRAC reasoning system (prac.core.base.PRAC) and a list of natural-language sentences subject to inference.

The method next_module returns the name of the next module to be executed according to the predefined PRAC pipeline. An instance of this module can then be created and executed:

>>> modulename = inference.next_module()
>>> module = prac.module(modulename)
>>> prac.run(inference, module)

PRACInference.next_module()¶

PRACInferenceStep¶

A PRACInferenceStep instance encapsulates a single inference step in the PRAC pipeline. It consists of n input databases and m output databases. Initialize an instance of the class prac.core.inference.PRACInference with a reference to the PRACInference object, a reference to the PRACModule performing this inference step and list of databases taken as inputs.

WordNet¶

The class prac.core.wordnet.WordNet is a wrapper class for Wordnet, which may be initialized with some WordNet concepts spanning an initial, collapsed concept taxonomy. The methods synsets and synset allow creating WordNet synsets from strings and perform a number of preprocessing steps to generate temporary WordnetConcepts for numbers that are not contained in the WordNet ontology. The method similarity will calculate the WUP- or path-similarity provided by WordNet after preprocessing the synsets. In particular, if at least one of the synsets is an adjective, this preprocessing includes looking up the synset in predefined similarity tables for colors, shapes and sizes and numbers not contained in the original ontology. Otherwise the derivationally related noun form of the adjective is retrieved from WordNet and used for the similarity calculation.

WordNet.synsets(*args, **kw)¶

WordNet.synset(*args, **kw)¶

WordNet.similarity(*args, **kw)¶

Reasoning¶

Once a PRAC- and a PRACInference instance have been initialized, we can run the PRAC pipeline.

>>> prac = PRAC()
>>> inference = PRACInference(prac, sentences)
>>> while inference.next_module() != None and not is_inference_process_aborted:
>>>     modulename = inference.next_module()
>>>     module = prac.module(modulename)
>>>     prac.run(inference, module)

>>> prac = PRAC()
>>> inference = PRACInference(prac, sentences)
>>> while inference.next_module() != None and not is_inference_process_aborted:
>>>     modulename = inference.next_module()
>>>     module = prac.module(modulename)
>>>     prac.run(inference, module)

>>> prac = PRAC()
>>> inference = PRACInference(prac, sentences)
>>> while inference.next_module() != None and not is_inference_process_aborted:
>>>     modulename = inference.next_module()
>>>     module = prac.module(modulename)
>>>     prac.run(inference, module)

API Reference¶

class prac.core.base.PRAC(configfile=’pracconf’)¶

The PRAC reasoning system.

construct_global_mln()¶: Reads all predicte declaration MLNs of all modules and returns an MLN with all predicates declared.

module(modulename)¶: Returns a loaded and initialized module given by the module name.

modules()¶: Returns a generator iterating over all module manifests.

query(instr, stopat=None)¶

Performs a query on PRAC using the natural-language instruction(s) instr.

Parameters:	instr – (str/list) A string or a list of strings representing an instruction or a list of instuctions. stopat – the name of the reasoning module where the pipeline is supposed to stop.

tell(howto, steps)¶

This method tells PRAC how complex high-level tasks are being achieved by executing multiple instruction steps.

This method triggers a (partial) PRAC inference pipeline and stores the result in the PRAC MongoDB.

Parameters:	howto – (str) a natural-language instruction describing the high-level task, e.g. ‘Make pancakes’ steps – list of natural-language instructions that achieve the high-level goal, e.g. [‘flip the pancake around.’, ‘wait for 2 minutes.’, …]

training_dbs(actioncore_name=None)¶: Returns the list of training database file names associated to the given action core. Returns all training databases if actioncore_name is None.

class prac.core.base.PRACModule(prac)¶

Base class for all PRAC reasoning modules. Provides some basic functionality for serializing, deserializing and running PRAC modules. Every PRAC module must subclass this.

static from_manifest(manifest, prac)¶: Loads a Module from a given manifest. :param manifest: a PRACModuleManifest instance :param prac: a PRAC instance :return: an instance of PRACModule

initialize()¶: Called after the PRAC module has been loaded. Every PRAC module can do some initialization stuff in here. The default implementation does nothing.

mlnquery(config=None, verbose=None, **params)¶

Wrapper for MLNQuery to replace the resultdb of the inference object with an MLN Database casted to a PRACDatabase

Parameters:	config – the configuration file for the inference verbose – boolean value whether verbosity logs will be printed or not params – dictionary of additional settings
Returns:	the inference object

shutdown()¶: Called when the PRAC reasoning system is to be shut down. Here modules can do some cleaning up. The default does nothing.

class prac.core.wordnet.WordNet(*args, **kw)¶

Wrapper class for WordNet, which may be initialized with some WordNet concepts spanning an initial, collapsed concept taxonomy.

Also provides a set of customized similarity measures for colors, shapes and sizes as introduced in Mareike’s thesis.

nltkpos(penntreepos)¶: Converts a Penn Treebank part-of-speech tag into an NLTK POS tag.

class prac.core.inference.PRACInference(prac, instr)¶: Represents an inference chain in PRAC

class prac.core.inference.PRACInferenceStep(node, module)¶: Wrapper class encapsulating a single inference step in the PRAC pipeline. It consists of n input databases and m output databases.