Skip to content

Latest commit

 

History

History
48 lines (34 loc) · 3.02 KB

prolog.md

File metadata and controls

48 lines (34 loc) · 3.02 KB
Raw

Aplib's Prolog Backend

aplib provides a Prolog backend, namely tuprolog, that your agent can use to do reasoning. The access to this backend is through the class nl.uu.cs.aplib.Agents.StateWithProlog whose rootclass is nl.uu.cs.aplib.mainConcepts.SimpleState. Recall that an instance of SimpleState represents an agent state. Any agent would need a state to be attached to it, and it should be an instance of SimpleState, as in:

var state = new StateWithProlog() ;
var agent = new BasicAgent() . attachState(state)

An instance of StateWithProlog holds an instance of Prolog-engine. This engine is hidden from you, but when programming actions for your agents, you can access it through methods state.addFacts(..) and state.addRules(..) to add facts and reasoning rules to the Prolog-engine. You can subsequently use the methods state.query(..) and state.test(...) to query the Prolog-engine.

A Prolog-engine is useful when your have a non-trivial state structure and some actions of your agent need to inspect if a certain complex predicate P(X,Y,Z) has a solution in the agent's current state (instances for X,Y,Z that would make P true). Using Prolog allows you to declaratively describe how P can be inferred using one ore more inference rules, and leave it to a Prolog-engine to do the hard work of search for its solution. To be able to make use of this, you of course will need to formulate the inference rules. Additionally, you need to encode fragments of your agent's state that would be relevant for the reasoning as Prolog facts and send these facts to the Prolog-engine.

There is an example of using Prolog in the class nl.uu.cs.aplib.exampleUsages.fiveGame.FiveGame_withAgent. This class contains an aplib agent for playing a simple board game called Five Game. It is played on a board of N x N by two players, each takes turn to place a piece of their respective color on a still empty square on the board. The first player that can form a consecutive row or column or diagonal consisting of 5 pieces all of his color wins. The agent player is programmed to use some bits of Prolog reasoning as part of its tactic.

For example, to add reasoning rules we do the following:

var state = new StateWithProlog() ;
state.addRules(
   "bossOf(A,C) :- bossOf(A,B), bossOf(B,C)",
   "isBoss(A) :- bossOf(A,B)"
)

Suppose now we add the following facts to the Prolog-engine:

state.addFacts(
   "bossOf(mrcrab,bob)",
   "bossOf(bob,patrick)"
)

Now we can query the Prolog-engine, e.g. to check if patrick is a boss:

state.test("isBoss(patrick)") // should return false

We can query the engine to give us a boss, if any:

var solution = state.query("isBoss(X)") ;
var boss = solution.str_("X") ; // should contain "mrcrab"

The class StateWithProlog is btw also a subclass of StateWithMessenger. So it also gives you access to inter-agent messaging.