pakozm / lua-faces

Lua-FaCES is a forward chaining expert system written in Lua programming language

Lua-FaCES

Lua-FaCES is a forward chaining expert system development tool written in Lua programming language.

Installation

Just copy faces.lua and tuple.lua to your Lua library location, usually in Ubuntu /usr/share/lua/5.2.

Tutorial

Once it is installed you can use the library by requiring it and instantiating a knowledge-base:

> faces = require "faces"
> kb = faces()

The knowledge-based (KB) is a set of facts which are known as true, and a set of rules which allow to update the facts. The most basic operation is definition of initial facts using function kb:fassert{...} which receive a table with the fact definition. A fact can be described as a tuple of strings, numbers or other tuples, so you can define a hierarchy of symbols. Every tuple and its compounds is considered a symbol. The following example asserts in the KB a fact describing a color as green:

> = kb:fassert{ "color", "green" }
2

The number returned by the function is a numeric identifier which allow you to retract or point this fact in different places. Using kb:facts() you can see the list of facts until now:

> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-2	tuple{ "color", "green" }
# For a total of 2 facts

As you can see facts are stored as tuples, so they are immutable and interned by the library (similar to Python tuples). At this point you can assert other facts.

> kb:fassert{ "color", "green" }
2
> kb:fassert{ "color", "red" }
3
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-2	tuple{ "color", "green" }
f-3	tuple{ "color", "red" }
# For a total of 3 facts

And it is possible to use kb:retract(...) to remove any of this facts:

> kb:retract(2)
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-3	tuple{ "color", "red" }
# For a total of 2 facts

The index given to retract is the fact index returned by kb:fassert(...) function. As any other Lua value you can store this index in a variable, operate with it, or whatever you want.

Facts are very limited, so rules are needed to allow transformation of facts, and finally, to allow computations. The following rule declares that if an animal is a duck then it should sound quack:

> kb:defrule("duck"):pattern{ "animal_is", "duck" }:
  ENTAILS("=>"):
  fassert{ "sound_is", "quack" }

The rule consists in three parts, a name (in this case "duck"), the left-hand-side (LHS) of the rule where pattern matching is used to state premises of the rule, and the right-hand-side (RHS) where actions are executed once the premises are stated as true.

At any moment you can see the list of available rules using kb:rules():

> kb:rules()
# Rules list
duck	salience:	0

In order to see this rule working, you need to assert the fact {animal_is duck}, so proceed this way and see what happens:

> kb:agenda()
# Agenda
# For a total of 0 activations
> = kb:fassert{ "animal_is", "duck" }
4
> kb:agenda()
# Agenda
0	duck	tuple{ 4 }
# For a total of 1 activations
> kb:run()
> kb:agenda()
# Agenda
# For a total of 0 activations
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-3	tuple{ "color", "red" }
f-4	tuple{ "animal_is", "duck" }
f-5	tuple{ "sound_is", "quack" }
# For a total of 4 facts

We have been used kb:agenda() in order to check which rules have their premises satisfied and which facts are satisfying its premises. Additionally, a number indicating the salience (priority) of the rule is shown. Higher salience value mean higher priority.

Now we are going to chain two rules in order to produce our expected goal of {sound_is duck}. First, using kb:clear() we remove all asserted facts, and re-asserting a new {initial fact}. After that, we define a new rule which declares the animal is a duck if it has webbed-feet and feathers:

> kb:clear()
> kb:defrule("is_it_a_duck"):
    pattern{"animal_has", "webbed_feet"}:
    pattern{"animal_has", "feathers"}:
    ENTAILS("=>"):
    fassert{"animal_is", "duck"}

Now assert both facts animal has and run the inference:

> kb:fassert{"animal_has", "webbed_feet"}
> kb:fassert{"animal_has", "feathers"}
> kb:run()
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-2	tuple{ "animal_has", "webbed_feet" }
f-3	tuple{ "animal_has", "feathers" }
f-4	tuple{ "animal_is", "duck" }
f-5	tuple{ "sound_is", "quack" }
# For a total of 5 facts

As you can see we end with five facts, the initial one, the two we have asserted, and the two asserted by the rules. The system executes first is_it_a_duck rule and after duck rule. Chaining rules in different ways you can describe really complex systems.

Working with variables and user defined functions

The pattern matching algorithm used by Lua-FaCES allow to declare variables in rules which value is instantiated to fact contents. As example, the following rule is executed with every {animal_has ...} fact and prints them into screen:

> kb:defrule("debug"):
    pattern{ "animal_has", "?x" }:
    ENTAILS("=>"):
    u(function(vars) print(vars.x) end)
> kb:run()
webbed_feet
feathers

As you can see, we have introduced here two concepts, variables and user defined functions. Variables are a string which starting with a question mark ?name. Once the variable is declared, you can use it without the question mark inside user defined functions. The user defined functions receive one and only one argument, which a list of all the variables declared in the rule. User defined functions can be expressed as a Lua function which receives the vars argument, or in a more compact way using a string. The string should be a Lua expression which can be evaluated as result of the function. This user defined functions can be declared at LHS as predicates, and at RHS as actions, so you can define powerful rules like the following one:

> kb:defrule("factorial"):
    pattern{ "Factorial", "?x", "?y" }:
    pattern{ "Limit", "?z" }:
    u"x < z":
    ENTAILS"=>":
    fassert{ "Factorial", u"x+1", u"y*(x+1)" }

This rule computes factorial numbers until a given limit number. It needs a base case {"Factorial",1,1} to be asserted and a {"Limit",10} for example:

> kb:clear()
> kb:fassert{"Factorial",1,1}
> kb:fassert{"Limit",10}
> kb:run()
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-2	tuple{ "Factorial", 1, 1 }
f-3	tuple{ "Limit", 10 }
f-4	tuple{ "Factorial", 2, 2 }
f-5	tuple{ "Factorial", 3, 6 }
f-6	tuple{ "Factorial", 4, 24 }
f-7	tuple{ "Factorial", 5, 120 }
f-8	tuple{ "Factorial", 6, 720 }
f-9	tuple{ "Factorial", 7, 5040 }
f-10	tuple{ "Factorial", 8, 40320 }
f-11	tuple{ "Factorial", 9, 362880 }
f-12	tuple{ "Factorial", 10, 3628800 }
# For a total of 12 facts

Fact identifiers can be captured in order to allow references to them into RHS part of the rule:

> kb:defrule("retract_all"):
    var("?f"):pattern{ "Limit", 10 }:
    ENTAILS("=>"):
    retract(u("f")):
    fassert{ "Limit", 12 }
> kb:run()
> kb:facts()
# Facts list
f-1	tuple{ "initial fact" }
f-2	tuple{ "Factorial", 1, 1 }
f-4	tuple{ "Limit", 12 }
f-5	tuple{ "Factorial", 2, 2 }
f-6	tuple{ "Factorial", 3, 6 }
f-7	tuple{ "Factorial", 4, 24 }
f-8	tuple{ "Factorial", 5, 120 }
f-9	tuple{ "Factorial", 6, 720 }
f-10	tuple{ "Factorial", 7, 5040 }
f-11	tuple{ "Factorial", 8, 40320 }
f-12	tuple{ "Factorial", 9, 362880 }
f-13	tuple{ "Factorial", 10, 3628800 }
f-14	tuple{ "Factorial", 11, 39916800 }
f-15	tuple{ "Factorial", 12, 479001600 }
# For a total of 14 facts

Multi-valuated variables

Lua-FaCES does not allow general multi-valuated variables, but it uses a simple trick which allow to solve the majority of cases. For example, let see the following facts:

> kb:clear()
> kb:fassert{ "A", { "multi-valuated", "example" } }
> kb:fassert{ "Another", { "one", "multi-valuated", "example" } }

Both facts are declared as a hierarchy of tuples, the fact tuple and a nested tuple at position 2. This kind of nested tuples can be captured as multi-valuated variables as follows:

> kb:defrule("MultiValuated"):
    pattern{ "?x", "$?y" }:
    ENTAILS("=>"):
    u(function(vars) print("x="..vars.x, "y="..tostring(vars.y)) end)
> kb:run()
x=A	y=tuple{ "multi_valuated", "example" }
x=Another	y=tuple{ "one", "multi_valuated", "example" }

As you can see, the rule is instantiated with both multi-valuated facts, so x variable contains the string at fact's head, and y variable is a tuple with the multi-valuated content. This multi-valuated variables only can be instantiated with tuples, so you need to take that into account when designing your KB. A tuple can be accessed as a plain Lua table, it allows pairs(), # and [] operators.