Goto Chapter: Top 1 2 3 4 5 Bib Ind
 [Top of Book]  [Contents]   [Previous Chapter]   [Next Chapter] 

3 Method Selection
 3.1 What are methods?
 3.2 How methods are called

3 Method Selection

The setup described in this chapter is intended for situations, in which lots of different methods are available to fulfill a certain task, but in which it is not possible in the beginning to decide, which one to use. Therefore this setup regulates, rather than just which method to choose, in which order the various methods are tried. The methods themselves return whether they were successful and, if not, whether it is sensible to try them again at a later stage.

The design is intentionally kept as simple as possible and at the same time as versatile as possible, thereby providing a useful framework for many situations as described above.

Note the differences to the GAP method selection, which is designed with the idea in mind that it will be quite clear in most situations, which one is "the best" method for a given set of input data, and that we do not want to try different things. On the other hand, the GAP method selection is quite complicated, which is to some extend necessary to make sure, that lots of different information about the objects in question can be used to really find the best method.

Our setup here in particular has to fulfill the requirement, that in the end, with lots of methods installed, one still has to be able to have an overview and to "prove", that the whole system always does the right thing.

3.1 What are methods?

A method is just a GAP function together with an agreement about what arguments it takes and what result it returns. The agreement about the arguments of course has to be made for every situation in which this generic method selection code is used, and the user is completely free there. A method can (and has to) return one of the following four values:

true

means that the method was successful and no more methods have to be tried.

false

means that the method was not successful and that there is no point to call the method again in this situation whatsoever.

fail

means that the method temporarily failed, that it however could be sensible to call it again in this situation at a later stage. This value is typical for a Las Vegas algorithm using randomised methods, which has failed, but which may succeed when called again.

NotApplicable

means that the method for some reason refused to do its work. However, it is possible that it will become applicable later such that it makes sense to call it again, may when more information is available.

For administration in the method selection, a method is described by a record with the following components bound:

method

holds the function itself.

rank

holds an integer used to sort the various methods. Higher numbers mean that the method is tried earlier. The numbering scheme is left to the user.

stamp

holds a string value that uniquely describes the method. This is used for bookkeeping and to keep track of what has to be tried how often.

comment

a string valued comment. This field is optional and can be left out.

The different methods for a certain task are collected in so-called "method databases". A method database is just a list of records, each describing a method in the format described above. Usually, the ranks will be descending, but that is not necessary.

There is one convenience function to put a new method into a method database:

3.1-1 AddMethod
‣ AddMethod( db, meth, rank, stamp[, comment] )( function )

Returns: nothing

db must be a method database (list of records, see above) with non-ascending rank values. meth is the method function, rank the rank and stamp a string valued stamp. The optional argument comment can be a string comment. The record describing the method is created and inserted at the correct position in the method database. Nothing is returned.

3.2 How methods are called

Whenever the method selection shall be used, one calls the following function:

3.2-1 CallMethods
‣ CallMethods( db, limit[, furtherargs] )( function )

Returns: a record ms describing this method selection procedure.

The argument db must be a method database in the sense of Section 3.1. limit must be a non-negative integer. furtherargs stands for an arbitrary number of additional arguments, which are handed down to the called methods. Of course they must fulfill the conventions defined for the methods in the database db.

The function first creates a "method selection" record keeping track of the things that happened during the method trying procedure, which is also used during this procedure. Then it calls methods with the algorithm described below and in the end returns the method selection record in its final state.

The method selection record has the following components:

falsemethods

a record, in which for every method that returned false the value 1 is bound to the component with name the stamp of the method.

failedmethods

a record, in which for every time a method returned fail the value bound to the component with name the stamp of the method is increased by 1 (not being bound means zero).

successmethod

the stamp of the method that succeeded, if one did. This component is only bound after successful completion.

result

a boolean value which is either true or fail depending on whether a successful method was found or the procedure gave up respectively. This component is only bound after completion of the method selection procedure.

tolerance

the number of times all methods failed until one succeeded. See below.

The algorithm used by CallMethods is extremely simple: It sets a counter tolerance to zero. The main loop starts at the beginning of the method database and runs through the methods in turn. Provided a method did not yet return false and did not yet return fail more than tolerance times before, it is tried. According to the value returned by the method, the following happens:

false

this is marked in the method selection record and the main loop starts again at the beginning of the method database.

fail

this is counted in the method selection record and the main loop starts again at the beginning of the method database.

NotApplicable

the main loop goes to the next method in the method database.

true

this is marked in the method selection record and the procedure returns successfully.

If the main loop reaches the end of the method database without calling a method (because all methods have already failed or are not applicable), then the counter tolerance is increased by one and everything starts all over again. This is repeated until tolerance is greater than the limit which is the second argument of CallMethods. The last value of the tolerance counter is returned in the component tolerance of the method selection record.

Note that the main loop starts again at the beginning of the method database after each failed method call! However, this does not lead to an infinite loop, because the failure is recorded in the method selection record such that the method is skipped until the tolerance increases. Once the tolerance has been increased methods having returned fail will be called again. The idea behind this approach is that even failed methods can collect additional information about the arguments changing them accordingly. This might give methods that come earlier and were not applicable up to now the opportunity to begin working. Therefore one can install very good methods that depend on some already known knowledge which will only be acquired during the method selection procedure by other methods, with a high rank.

 [Top of Book]  [Contents]   [Previous Chapter]   [Next Chapter] 
Goto Chapter: Top 1 2 3 4 5 Bib Ind

generated by GAPDoc2HTML