A toolkit for equivalence-based belief change
Equibel is a Python package for working with consistency-based belief change in a graph-oriented setting.
Currently Supported Platforms
- Mac OS X (tested on OS X 10.10.5) with Python 2.7.x
- 64-bit Linux (tested on Ubuntu 14.04) with Python 2.7.x
Note that while Equibel is distributed as a Python package, the core of the system is implemented using Answer Set Programming (ASP), and relies on an underlying ASP solver called clingo, which is part of the Potsdam Answer Set Solving Collection (Potassco).
In particular, Equibel has two ASP-related dependencies: the Python gringo module, which provides an interface to an ASP solver from within Python, and asprin.parser, which is a component of the asprin preference-handling framework. asprin is described in more detail here, and can be downloaded from here.
The Python component of Equibel is highly portable across platforms; however, the gringo and asprin.parser dependencies must be compiled for specific system configurations, producing system-specific binaries. In order to simplify usage for some common system configurations, Equibel includes pre-compiled binaries of these dependencies for 64-bit Ubuntu Linux and Mac OS.
If Equibel does not function correctly once it is installed, this is likely due to the fact that the pre-compiled binaries are not compatible with your system. In this case, you must compile the dependencies manually, by downloading the required components directly from Potassco. In order to use your custom binaries with Equibel, it is recommended that you follow the installation instructions given on the Github project page, which involve downloading the source code of Equibel to provide access to directories in which you can overwrite the included binaries with your own.
The following steps assume that you have the pip Python package manager installed. If you don’t have pip, you can get it here.
The pre-compiled gringo modules included with Equibel for either 64-bit Linux or Mac OS require a dependency called Threading Building Blocks (tbb).
The easiest way to install the tbb library on Mac OS is to use Homebrew:
$ brew install tbb
On Ubuntu Linux, the tbb library can be installed using the apt package manager:
$ sudo apt-get install libtbb-dev
Install Equibel using pip:
$ pip install equibel
If you are installing Equibel system-wide, you may need to use sudo:
$ sudo pip install equibel
To use Equibel within a Python program, you need to import the equibel package. The statement import equibel as eb imports this package, and gives it a shorter alias eb. The following Python script creates a path graph, assigns formulas to nodes, finds the global completion, and prints the resulting formulas at each node:
import equibel as eb if __name__ == '__main__': # Create an EquibelGraph object, which represents a graph and # associated scenario. G = eb.EquibelGraph() # Create nodes: G.add_nodes_from([1, 2, 3, 4]) # Create edges: G.add_edges_from([(1,2), (2,3), (3,4)]) # Add formulas to nodes: G.add_formula(1, "p & q & r") G.add_formula(4, "~p & ~q") # Find the global completion of the G-scenario: R = eb.global_completion(G) # Pretty-print the resulting formulas at each node: eb.print_formulas(R)
If the above code is saved in a file called completion.py, then it can be run by typing python completion.py at the command line, as follows:
$ python completion.py Node 1: p ∧ q ∧ r Node 2: r Node 3: r Node 4: r ∧ ¬p ∧ ¬q
If you get an error running this example, it is most likely due to the dependencies of Equibel not being compatible with your platform. As noted above, Equibel includes pre-compiled binaries of the Python gringo module, as well as of asprin.parser, for 64-bit Linux distributions (tested on Ubuntu 14.04) and for Mac OS (tested on OSX 10.10.5). If you are not using one of these systems, you will need to manually compile the gringo and asprin.parser dependencies (see the Github page).
Equibel allows for experimentation with several different approaches to consistency-based belief change in a graph-oriented setting, namely:
- Global completion
- Simple iteration
- Expanding iteration
- Augmenting iteration
- Ring iteration
The global completion operation is performed on an EquibelGraph G by eb.global_completion(G); this performs a “one-shot” procedure to update the information at every node in the graph, and thus is not an iterative approach. All of the other approaches—simple, expanding, augmenting, and ring—can be performed iteratively, and each one iterates to a fixpoint. The table below summarizes the Equibel functions used to perform single iterations of each approach, as well as to find the fixpoints reached by each approach:
|Method||Single Iteration||Iterate to Fixpoint|
Each of the approaches has two separate implementations, corresponding to its equivalent semantic and syntactic characterizations. In addition, there are two ways of performing the core optimization procedure over equivalences, involving either inclusion-based or cardinality-based maximization.
Each function listed above can take three optional arguments:
- method, which is a string that is either “semantic” or “syntactic”,
representing the method to use when performing the approach; e.g.
based on either the syntactic or semantic characterizations
- The default method is semantic
- To avoid typos when entering strings, Equibel has constants eb.SEMANTIC and eb.SYNTACTIC which equal the strings “semantic” and “syntactic”, respectively.
- opt_type, which is a string that is either “inclusion” or “cardinality”,
representing the type of maximization to be performed over equivalences
- The default opt_type is inclusion
- To avoid typos when entering strings, Equibel has constants eb.INCLUSION and eb.CARDINALITY which equal the strings “inclusion” and “cardinality”, respectively.
- simplify, which is a Boolean flag specifying whether to simplify the
final formulas at each node.
- The default value for simplify is False
By definition, the semantic and syntactic characterizations of an approach yield equivalent results; however, depending on the input scenario and type of approach, the performance of the characterizations may differ significantly. A good example of this is in the case of expanding iteration, where we have an early-stoppping condition over the radius of the expanding neighbourhood when using the semantic characterization, but not when using the syntactic characterization (causing the semantic characterization to be significantly faster for large graphs in practice).
To show how the method and opt_type arguments can be combined, we consider the following (by no means exhaustive) examples.
In the following example, we can see the difference between using inclusion-based optimization and cardinality-based optimization in the global completion:
import equibel as eb if __name__ == '__main__': # Creates a star graph with nodes [0, 1, 2, 3] and undirected edges [(0,1), (0,2), (0,3)] G = eb.star_graph(3) G.add_formula(1, 'p') G.add_formula(2, 'p') G.add_formula(3, '~p') # Using inclusion-based maximization over equivalences R_inclusion = eb.global_completion(G, method=eb.SEMANTIC, opt_type=eb.INCLUSION, simplify=False) eb.print_formulas(R_inclusion) # Using cardinality-based maximization over equivalences R_cardinality = eb.global_completion(G, method=eb.SEMANTIC, opt_type=eb.CARDINALITY, simplify=False) eb.print_formulas(R_cardinality)
Saving this code in a file inclusion_vs_cardinality.py and running it yields:
$ python inclusion_vs_cardinality.py Node 0: p ∨ ¬p Node 1: p Node 2: p Node 3: ¬p Node 0: p Node 1: p Node 2: p Node 3: ¬p
The following example function calls for the global completion operation show the flexible way in which options can be combined in Equibel:
- R_semantic = eb.global_completion(G)
- This function call computes the global completion of G. With no options explicitly specified, the defaults are used; thus, this call involves the semantic characterization with inclusion-based optimization, and does not simplify the resultant formulas.
- With all options explicitly specified, the above function call is equivalent to R_semantic = eb.global_completion(G, method=eb.SEMANTIC, opt_type=eb.INCLUSION, simplify=False)
- R_syntactic = eb.global_completion(G, method=eb.SYNTACTIC)
- This finds the global completion of G, using the syntactic characterization, the default inclusion-based optimization, and no simplification of formulas.
- R_syntactic = eb.global_completion(G, method=eb.SYNTACTIC, opt_type=CARDINALITY)
- This finds the global completion of G, using the syntactic characterization, cardinality-based optimization, and no simplification of formulas.
- R_syntactic = eb.global_completion(G, method=eb.SYNTACTIC, opt_type=CARDINALITY, simplify=True)
- This finds the global completion of G, using the syntactic characterization and cardinality-based optimization. With the simplify=True option, the resulting scenario will have simplified formulas for each node in the graph.
These options can be similarly combined for each of the iterative approaches, as shown in the following example calls:
- R_semantic = eb.iterate_simple(G, method=eb.SEMANTIC, simplify=True)
- This function call computes the graph and scenario that result from performing a single simple iteration over G, using the semantic characterization with default inclusion-based optimization. With the simplify=True option, the resulting scenario will have simplified formulas for each node in the graph.
- R_syntactic = eb.iterate_simple(G, method=eb.SYNTACTIC, simplify=True)
- This call is similar to the previous call, except that it uses the syntactic characterization of simple iteration, rather than the semantic characterization.
- R_semantic_fixpoint = eb.iterate_simple_fixpoint(G, method=eb.SEMANTIC, opt_type=eb.CARDINALITY, simplify=True)
- This computes the fixpoint reached by a sequence of simple iterations starting from the graph and scenario represented by G, using the semantic characterization and cardinality-based optimization.
- R_semantic = eb.iterate_expanding(G, simplify=True)
- This function call computes the graph and scenario that result from performing a single expanding iteration over G, using the default semantic characterization with default inclusion-based optimization. Since simplify=True, the resulting scenario will have simplified formulas for each node in the graph.
- R_semantic = eb.iterate_augmenting_fixpoint(G, simplify=True)
- This computes the fixpoint reached by a sequence of augmenting iterations starting from the graph and scenario represented by G, using the default semantic characterization and inclusion-based optimization. Since simplify=True, the resulting scenario will have simplified formulas for each node in the graph.