Artificial Intelligence

Artificial IntelligenceIntelligence

Ian [email protected]

Practical 2: Forward Checking

Artificial IntelligenceIntelligence

Part I : OverviewPart II: Three ways to implement FCPart III: Other parts of the practicalPart IV: What I’m looking for

Practical 2: Forward Checking

3

Practical 2: Forward Checking

Write a program to implement the two algorithms BT (Backtracking) and FC (Forward Checking.) Perform an empirical comparison of the two algorithms.

Some practical stuff: This is practical 2 of 2. Each will carry equal weight, I.e. 10% of total credit You may use any implementation language you wish Deadline(s) are negotiable (can be decided after vacation)

4

Aims and Objectives

Aims: to give experience in implementing important AI search

algorithms to give experience in comparing AI techniques empirically

Objectives: after completing the practical, you should have:

implemented the algorithms BT and FCgained an appreciation of some of the basic techniques necessaryperformed and reported on an empirical comparison of different

algorithms

5

What you need to do Implement BT and FC for binary CSP’s

if you can do FC you can do BT FC is the hard bit implement at least two (static) heuristics for each

Implement a reader to read in benchmark CSP’s format of problems will be provided use benchmarks for testing

Perform empirical comparison of algorithms run on benchmark problems report on comparative success of algorithm/heuristic

combinations

6

What you can get away with Implement BT binary CSP’s

implement at least one heuristics Implement a reader to read in benchmark CSP’s

format of problems will be provided use benchmarks for testing

Perform empirical comparison of algorithms run on benchmark problems report on success or otherwise

Don’t expect too many marks for doing the above but don’t expect zero either

7

Three Ways to Implement FC

You only need one implementation! Choose the style that suits you and the language

you like usingThree ways are:

using the general search algorithm recursive from pseudocode using specific data structures

8

Implementing FC (1) You can implement FC using the generic search algorithm presented

earlierSearch states = some representation of current assignment of values

to variables, and current domains for each variableForward checking done when new states createdDo search by depth-firstMain problem is memory management

not letting space expand endlessly/overwriting existing states easier if you’ve got GC built in

Appropriate for languages with non destructive data structures (e.g. Lisp, Haskell)

9

FC via general search algorithm1. Form a one element list with null state

• null state = state with no decisions = original CSP

2. Loop Until (either list empty or we have a solution) Remove the first state S from the list Choose the next decision to make

• which variable x to assign next Create a new state for each possible choice of decision

• decisions are all remaining values v in Dx

• to create each new state, assign x=v and forward check MERGE the set of new states into the list

3. If (solution in list) succeed and report solution else list must be empty, so fail

10

Implementing FC (2)

Functional languages are good for search e.g. Lisp, Haskell

Write propagator for forward checking which makes non destructive changes. I.e. original state still exists, but we get a new one for free GC done for you

Write search function recursively handles the manipulation of the list for you via the function

calling stack

11

Implementing FC (2)

Search (CSP): choose var while (value remains in CDvar)

Call Search( fc-propagate(CSP[var = value]))If call succeeds with solution, return solution

If all calls failed, return failure

12

Implementing FC(3)

Follow implementation outlined by ProsserAvoids most memory management problemsExplicit data structures initially set up

when we remove values from vi to vj we modify them reductions[j] contains sequence of sequence

each one a sequence of values disallowed by past var past-fc[j] is a set of variables

set of variables i which caused value removals from vj

future-fc[i] is another setset of variables in which the current value of vi causes value

removals

13

General pseudocode for bcsspProcedure bccsp (n, status)

consistent := true, status := unknown, ii := 1 while (status = unknown)

if (consistent) • ii := label(ii,consistent)

– need special purpose function fc-label here

• else ii := unlabel(ii,consistent)– and fc-unlabel here

if (ii > n)• status := solution• else if (ii = 0)

– status := impossible

14

Implementing FC(3.2)Use data structure suggested by Bacchus/van RunHave a 2D array Domain[ii,k]

first dimension is variables, second dimension values Domain[ii,k] = 0 if value k still possible for variable ii

I.e. if k still belongs to CD[ii] If value k impossible, removed from CD[ii]

Domain[ii,k] = j, where j is variable that caused removalOn backtracking, to undo effect of assigning j

if Domain[ii,k] = j, reset it so that Domain[ii,k] = 0 either store all changes made by j, or just iterate over 2D array looking for those equal to j

when we remove values from vi to vj we modify them reductions[j] contains sequence of sequence

each one a sequence of values disallowed by past var past-fc[j] is a set of variables

set of variables i which caused value removals from vj

future-fc[i] is another setset of variables in which the current value of vi causes value removals

15

Other parts of the practical

Input format: the APES group has a standard format for sharing binary

CSP’s. Allows sharing of benchmarks Valuable for testing (all programs should give same results)

Write a reader for this format translate input to your internal format for CSP

your representation of variables, domains, constraints create small test problems for yourself

and if you want, share them for others

16

Heuristics

I am only looking for static variable ordering heuristics implement dynamic ones if you wish heuristics are harder in Prosser’s version

see paper by Bacchus & van Run for pointersHeuristics you might consider

lexicographic, v1, v2, v3… random, v17, v16, v2, v19 … min degree: var involved in least constraints first max degree: var involved in most constraints first other heuristics you find/can think of

17

Empirical ReportRun your program(s) against benchmark instances I

will provide, and others you might want to tryFrom empirical evidence, how do the techniques

perform? Is FC better than BT? Worse? varies across problems? Are there some problems that you can’t solve in

reasonable cpu time? Is min degree better than max degree? Are some problems harder than others?

18

Empirical ReportWrite a report on your experimentsDescribe the purpose of each experiment, the

results, and conclusions you drawTry to make it a good piece of empirical AI! Include results as e.g. tables or graphs

as appendix if too many results Probably a few pages

19

What I am looking for A correct functioning program

speed is not important (within reason) should implement at least 4 combinations of

algorithm/heuristicA report summarising program and empirical work

no set word limit, probably needs a few pages to present good empirical work well

evidence that your code is correcte.g. sample output, correct result on benchmarks

conclusions on your empirical result code (electronically if it’s HUGE)

20

Additional Issues

Some ways to get more credit … create/find problems for which usually worse

algorithm/heuristic does better think of different heuristics think of interesting hypotheses and test them implement FC so that propagation causes a chain reaction.

I.e. if you get domain size = 1, redo FC from thereSince I’ve asked for static heuristics, we may search on

variable x, domain size 4, when variable y has d.s. = 1 implement dynamic variable ordering heuristics

21

Some pointers

A tutorial on constraint programming Barbara Smith Leeds University, 1995

Hybrid Algorithms for the Constraint Satisfaction Problem Patrick Prosser Computational Intelligence, 1993

Dynamic Variable Ordering in CSPs Bacchus & van Run CP95, 1995