July 2005


Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Sean Luke <[log in to unmask]>
Reply To:
ECJ Evolutionary Computation Toolkit <[log in to unmask]>
Wed, 27 Jul 2005 02:14:03 -0400
TEXT/PLAIN (92 lines)
On Jul 27, 2005, at 12:00 AM, Mozaherul Hoque Abul Hasanat wrote:

Is there any way to specify the root nodes of the GP individuals to be
a particular function, say .. Division or Multiplication?

Absolutely.  Here's an example params file.

###### example params file, goes in ec/app/regression
parent.0 = ./noerc.params

# To require that the root of a tree be a given node, you
# simply need to set up the typing so that that node is
# the only node which is compatable with the tree's return
# type.  The way to do this is to make not one but two
# atomic types (in the example below, nil and root).  nil
# is the default type used by our GP nodes.  root will be
# the return type of the tree.  Then we create a set-type
# which is compatable with both of them called nil-and-root,
# and set things up so the return type of our desired node
# (in this case, 'div'), returns nil-and-root.  Thus 'div'
# is legal to hang as root (and is the only such node), AND
# 'div' is still perfectly compatible as a child of other
# nodes as before (because it still meshes with nil).

# The gotcha: many GP node builders require that for every
# reachable type used by nodes in the function set there
# must exist at least one terminal (and ideally one nonterminal)
# in the function set which returns that type.  This is
# because many GP node builder algorithms insist on being
# able to emit a terminal whenever they so desire, regardless
# of what the current type situation is where the terminal
# has to be hung.  The notable exception to this is Uniform.

# We have only one node which returns nil-and-root,
# and it's a nonterminal ('div').  Is this going to trip us up?
# No.  Ordinarily it would be a problem for us as we use
# Ramped Half-and-Half as our builder, and it has this
# requirement.  However since we are telling Ramped Half-and-Half
# to generate only trees of sizes 2 and up (NOT 1 and up),
# it will never attempt to select a terminal for the root
# and this gotcha will not occur.  The issue would also come
# up if we use tree mutation as an operator rather than solely
# crossover.  But we're not doing that.

# If you want to see the problem occur, change Ramped Half-and-Half
# to generate trees of sizes 1 and up.  Then occasionally it'll try
# to make a tree that's just a terminal which returns root-and-nil,
# and there is no node that can fill that requirement.  Here's the
# parameter which will make that happen:
# gp.koza.half.min-depth = 1

# Okay, here we go.

# Add an atomic type (root) and a set type
# (nil-and-root) which is compatible with both
# nil and root
gp.type.a.size = 2 = nil = root
gp.type.s.size = 1 = nil-and-root
gp.type.s.0.size = 2
gp.type.s.0.member.0 = nil
gp.type.s.0.member.1 = root

# make a new contraints form for div which returns
# things of type nil-and-root.  This is compatable
# with all elements that accept nil (which will be
# all other functions), so div will still fit right in.
# Note that this constraints takes two kids just like
# div would expect. = 8 = = nc-returning-nil-and-root = nil-and-root = 2 = nil = nil

# redefine div to return things of type nil-and-root.
# (in erc.params, div is function 4) = nc-returning-nil-and-root

# define the tree so that it only accepts items that
# are compatible with root.  div will be the only
# such item. = root

###### end example