Abnormal Predicates: Learning Categorical Defaults from Probabilistic Rules MSc Project - Rose Azad Khan ======================== This repository contains the original ProbFOIL code, along with several files and folders created for use in this MSc project.

In the probfoil folder: every file in this folder belongs to the original ProbFOIL implementation, except for defaults.py, defaults_complex.py and generator.py. The original ProbFOIL files are unmodified except for probfoil.py

defaults.py contains the function construct_ab_pred, which creates the abnormal predicate and new data points and writes these to the data files.

defaults_complex.py is identical to defaults.py except that it includes a loop which activates when there is more than one positive predicate in the body of the rule i.e. more than one candidate for creating ab_pred. This file was used when complex examples were explored in the qualitative evaluation

generator.py is the program used to generate datasets for the quantitative evaluation. It can be run from the command line, you need to specify an output data file <file.data> and an output settings file <file.settings.pl> and the number of objects in the domain N

probfoil.py has been modified to learn categorical defaults rather than statistical defaults. At the moment, probfoil.py uses the function from defaults_complex.py (line 459 of probfoil.py). If you wish to use the function from defaults.py instead, you must change the import statement at the top of the file to 'import defaults' instead of import defaults_complex. The function names are identical

To run the modified probfoil algorithm, you need to use the repository version of probfoil. If you are using files from a separate folder, you will need to run probfoil as a module, due to importing The command is 'python -m probfoil.probfoil -p 0.0 -v <file.settings.pl> <file.data> ' The significance value p must be set to 0.0. The verbose setting is optional (-v, -vv, -vvv etc)

The folder 'datasets' contains the datasets which were generated for the quantitative evaluation, along with a file analysis.py which was used to plot the quantitative results on graphs

The folder 'examples' contains a large number of data files and settings file, which were used for experimentation and for learning complex examples.

The folder 'alchemy-files' contains example Alchemy .mln and .db files, from the initial experimentation with Alchemy

Running the modified probfoil algorithm writes data points to both the data file and the settings file. If you wish to run the algorithm more than once using the same files, you will need to open the files and delete the new data points, unless you wish to keep the new abnormal predicate for the next learning stage

The text below is the original ProbFOIL readme

ProbFOIL v2.1

ProbFOIL is a probabilistic extension of FOIL that is capable of learning probabilistic rules from probabilistic data.

ProbFOIL 2.1 is a redesign of the Prob2FOIL algorithm that was introduced in https://lirias.kuleuven.be/handle/123456789/499989. It works on top of ProbLog 2.1

If you are looking for the version used in the paper, you should check out the tag paper_version.

Installation

ProbFOIL 2.1 requires ProbLog 2.1. You can install ProbLog by using the command:

pip install problog

ProbFOIL does not require any further installation.

Usage

The input of ProbFOIL consists of two parts: settings and data. These are both specified in Prolog (or ProbLog) files, and they can be combined into one.

The data consists of (probabilistic) facts. The settings define

• target: the predicate we want to learn
• modes: which predicates can be added to the rules
• types: type information for the predicates
• other settings related to the data

To use:

probfoil data.pl

or, in the repository version

python probfoil/probfoil.py data.pl

Multiple files can be specified and the information in them is concatenated. (For example, it is advisable to separate settings from data).

Several command line arguments are available. Use --help to get more information.

Settings format

Target

The target should be specified by adding a fact learn(predicate/arity).

Modes

The modes should be specified by adding facts of the form mode(predicate(mode1, mode2, ...), where modeX is the mode specifier for argument X. Possible mode specifiers are:

• +: the variable at this position must already exist when the literal is added
• -: the variable at this position does not exist yet in the rule (note that this is stricter than usual)
• c: a constant should be introduced here; possible value are derived automatically from the data

Types

For each relevant predicate (target and modes) there should be a type specifier. This specifier is of the form base(predicate(type1, type2, ...), where typeX is a type identifier. Type can be identified by arbitrary Prolog atoms (e.g. person, a, etc.)

Example generation

By default, examples are generated by quering the data for the target predicate. Negative examples can be specified by adding zero-probability facts, e.g.:

0.0::grandmother(john, mary).

Alternatively, ProbFOIL can derive negative examples automatically by taking combinations of possible values for the target arguments. Note that this can lead to a combinatorial explosion. To enable this behavior, you can specify the fact

example_mode(auto).

Example

% Modes
mode(male(+)).
mode(parent(+,+)).
mode(parent(+,-)).
mode(parent(-,+)).

% Type definitions
base(parent(person,person)).
base(male(person)).
base(female(person)).
base(mother(person,person)).
base(grandmother(person,person)).
base(father(person,person)).
base(male_ancestor(person,person)).
base(female_ancestor(person,person)).

% Target
learn(grandmother/2).

% How to generate negative examples
example_mode(auto).

Further examples can be found in the directory examples.