Skip to content

Latest commit

 

History

History
369 lines (256 loc) · 11.2 KB

README.md

File metadata and controls

369 lines (256 loc) · 11.2 KB
Raw

pylo is a Python front-end for several logic programming engines. This includes several Prolog engines, but also different flavours of logic programming such as relational programming (kanren) and Datalog. It allows you to write your program once and execute it with different engines simply by switching the back-end.

Supported OS: The library was tested on Linux (Ubuntu) and OSX.

Supported engines

Prolog engines

Currently supported:

  • SWIPL
  • GNU Prolog (works only on OSX so far; GNU PROLOG's foreign function interface does not compile properly on Linux)
  • XSB Prolog

Under development:

Maybe supported in the future:

Datalog engines

A subset of Prolog without functors/structures

Currently supported:

Considering:

Relational programming engines

Prolog without side-effects (cut and so on)

Currently supported:

Deductive database engines

Currently supported:

  • none yet

Considering:

Answer set programming

Currently supported:

  • none yet

Considering:

Installation

Installation with pip

STEP 1: install you favourite Prolog engine(s). Make sure they are available through the terminal. pylo relies on the which command to identify the right paths.

STEP 2: clone this repository

STEP 3: install with pip

pip install -v .

STEP 4 (optional): Test

from pylo.tests import all_swipl_tests, all_gnu_tests, all_xsb_tests, test_kanren, test_datalog

all_swipl_tests()
all_gnu_tests()
all_xsb_tests("[path_to_XSB_folder]")  # the same folder for installation
test_kanren()
test_datalog()

If something goes wrong, refer to the INSTALL.md file for the alternative procedures.

Other practical considerations

  • if you are using zsh shell instead of bash, XSB will not be automatically detectable. You will have to manually set the right environmental variable pointing to the location of XSB sources, as described in INSTALL.md.

  • In order to use SWIPL, you need to provide a path to the dynamic library using the LD_LIBRARY_PATH env var. For instance, on OSX

   export LD_LIBRARY_PATH=/usr/local/Cellar/swi-prolog/8.2.0/libexec/lib/swipl/lib/x86_64-darwin
  • For Ciao prolog, you also need to be able to find the dynamic library. On OSX
    export LD_LIBRARY_PATH=/Users/seb/.ciaoroot/master/build/eng/ciaoengine/objs/DARWINx86_64/
  • Z3Py scripts stored in arbitrary directories can be executed if the 'build/python' directory is added to the PYTHONPATH environment variable and the 'build' directory is added to the DYLD_LIBRARY_PATH environment variable.

Usage

Language design

pylo follows the logic programming terminology, instead of Prolog terminology, The basic constructs include terms:

  • constants: luke, leia, anakin or Python numbers like 1, -3, 3.14
  • variables: X, Y
  • structures: sabre(green,long), date(2020,january,1)
  • lists (a special kind of structure): List([1,2,3,4,5])
  • pairs (a list made of head and tail): Pair(X,Y)

To construct clauses, one needs:

  • predicates: parent, ...
  • literals:
    • atoms (predicate applied to terms): parent(vader,luke), mother(padme, leia)
    • negations of atoms: Not(parent(vader,luke))
  • clauses: parent(X,Y) :- mother(X,Y)

To reduce the memory usage, constants, variables, functors and predicates should be constructed through context:

  • constants: c_const('luke') creates a constant luke (for numbers, just use Python structures)
  • variables: c_var('X') creates a variables X
  • predicates: c_pred('parent', 2) creates a predicate parent with arity 2
  • functors: c_functor('date', 3) creates a functor date with arity 3

If you want to use a proposition in a clause (a functor without arguments), like a :- ..., create it as a Predicate with arity 0. Be careful to construct clauses with literals, not structures.

Specifying knowledge

Pylo allows you to conveniently specify the knowledge base and the query it with different prolog engines. All basic constructs (constants, variables, functors and predicates) should be created using the global context (functions prefixed with c_: c_const, c_pred, c_var, c_functor), which ensures that there are not duplicates.

from pylo.language.lp import c_pred, c_var, c_const, c_functor, Atom, Clause, Body, Structure, List

# create some constants 
luke = c_const("luke")             
anakin = c_const("anakin")
leia = c_const("leia")
padme = c_const("padme")


# create predicates
father = c_pred("father", 2)      
mother = c_pred("mother", 2)
parent = c_pred("parent", 2)


# create literals
f1 = Atom(father, [anakin, luke])                    
f2 = Atom(father, [anakin, leia])
f3 = Atom(mother, [padme, luke])
f4 = Atom(mother, [padme, leia])


# create Variables
X = c_var("X")                   
Y = c_var("Y")


# create clauses 
head = Atom(parent, [X,Y])                           
body1 = Atom(father, [X,Y])
body2 = Atom(mother, [X,Y])
rule1 = Clause(head, Body(body1))
rule2 = Clause(head, Body(body2))


# create structures
sabre = c_const("sabre")
green = c_const("green")
item = c_functor("item")

struct = Structure(item, [sabre, green])


# create lists
l = List([padme, leia, 1, 2, 3])

Such an explicit object-oriented way might be suited for automated construction of programs. Pylo also provides many convenient shortcuts for less tedious construction of knowledge bases. The above example could have been constructed in the following way

from pylo.language.lp import c_pred, c_const, c_var, c_functor, List, Pair

# construct predicates
father = c_pred("father", 2)
mother = c_pred("mother", 2)
parent = c_pred("parent", 2)
belongs_to = c_pred("belongs_to", 2)

# create facts directly
#    applying the predicate symbol to terms/strings creates a literal
#        it also converts strings to the appropriate structures
f1 = father("anakin", "luke")
f2 = father("anakin", "leia")
f3 = mother("padme", "luke")
f4 = mother("padme", "leia")

# create structures directly
#     applying the functor to a series of terms/strings creates a structure
#     it also knows how to convert strings to the right form
item = c_functor("item", 2)
struct = item("sabre", "green")

f5 = belongs_to(struct, "luke")


# create list
l = List(["padme", "leia", 1, 2, 3])  # if constants are numbers, just use python data structures

# create pair
p = Pair("X", "Y")


# creating clauses
rule1 = parent("X", "Y") <= father("X", "Y")
rule2 = parent("X", "Y") <= mother("X", "Y")

Querying Prolog

The first step is to create a Prolog instance

from pylo.engines.prolog import SWIProlog
from pylo.engines.prolog import XSBProlog
from pylo.engines.prolog import GNUProlog


# Create GNU Prolog instance
pl_gnu = GNUProlog()

# create XSB Prolog
# you need to provide the path
pl_xsb = XSBProlog("[path to the XSB folder used in installation]")

# create SWI Prolog
# the path to SWIPL binary is optional if it corresponds
pl_swi = SWIProlog('/usr/local/bin/swipl') 

# create datalog instance
from pylo.engines.datalog import MuZ
solver_data = MuZ()

# create kanren instance
from pylo.engines.kanren import MiniKanren
solver_kan = MiniKanren()

Two things need to be noted:

  • SWIProlog takes the path to the executable as an argument. If the path is standard, i.e. is /usr/local/bin/swipl on OSX, no argument needs to be passed
  • XSBProlog takes a path to the XSB sources as an argument

All Prolog engines have a unified interface:

from pylo.engines.prolog import Prolog
pl = Prolog()

# consult file
pl.consult("metagol.pl")

# load module
# Pylo does not provide a standardization of modules accross different engines
#    this means that it is your job to ensure that the correct mofule nam is provided 
#
#    for instance, GNU prolog does not have modules
#                  XSB Prolog needs to specify which predicates to load frm the library, 
#                               this is provided as an extra names arguments 'predicates', which takes a list of predicates
pl.use_module("library(lists)")

# asserta a fact or a clause
pl.asserta()

# assertz a fact or a clause
pl.assertz()


# retract literal
pl.retract()

# check whether it is possible to satisfy a query
# has_solution(...) can take any number of literals as an argument, which is interpreted as a conjunction
pl.has_solution()

# query all solutions
# query(...) takes any number of literals as an input, which is interpreted as a conjuncion
#       it additionally takes 'max_solutions' arguments, which can be used to limit the number of solutions to look for
pl.query()

A more elaborate example

from pylo.engines.prolog import XSBProlog
from pylo.language.lp import c_pred 

pl = XSBProlog("/Users/seb/Documents/programs/XSB")

person = c_pred("person", 1)
friends = c_pred("friends", 2)
stress = c_pred("stress", 1)
influences = c_pred("influences", 2)
smokes = c_pred("smokes", 1)
asthma = c_pred("asthma", 1)

pl.assertz(person("a"))
pl.assertz(person("b"))
pl.assertz(person("c"))
pl.assertz(friends("a", "b"))
pl.assertz(friends("a", "c"))

pl.assertz(stress("X") <= person("X"))
pl.assertz(influences("X", "Y") <= person("X") & person("Y"))
pl.assertz(smokes("X") <= stress("X"))
pl.assertz(smokes("X") <= friends("X", "Y") & influences("Y", "X") & smokes("Y"))
pl.assertz(asthma("X") <= smokes("X"))

query_p = person("X")
tv = pl.query(query_p)
print("all persons: ", tv)

query_f = friends("X", "Y")
tv = pl.query(query_f)
print("all friends: ", tv)

query_st = stress("Y")
tv = pl.query(query_st)
print("all stressed people: ", tv)

tv = pl.query(influences("X", "Y"))
print("all influences: ", tv)

tv = pl.query(smokes("X"))
print("all smokers: ", tv)

tv = pl.query(asthma("X"), max_solutions=3)
print("all asthma: ", tv)

Missing features

  • providing python functions as predicates
  • full support for pairs ([|]/2 or ./2)
    • SWIPL (supports pairs)
    • XSB
    • GNU Prolog