Second Order Heuristics in ACGP
-
Upload
hauschildm -
Category
Technology
-
view
1.171 -
download
0
Transcript of Second Order Heuristics in ACGP
Second Order Heuristics in
ACGP
John AleshunasUniversity of Missouri –
Saint [email protected]
Cezary JanikowUniversity of Missouri –
Saint [email protected]
Mark W HauschildUniversity of Missouri –
Saint [email protected]
OutlineThe Genetic Programming (GP) ModelIssues with the GP ModelConstrained Genetic Programming (CGP)Adaptable Constrained GP (ACGP)Adaptable Constrained GP (ACGP)
HeuristicsACGP Learning ResultsConclusions
2
The Genetic Programming (GP) Model
Evolve a solution program from a population of candidates
Represents individual solutions as variable size trees
Employs standard genetic operators Selection Crossover Mutation
3
Issues with the GP Model
The Function set and Terminal set (labels) define the representation space of a GP implementation
Often the Function set and Terminal set are over specified to ensure that they cover the solution space
Closure Sufficiency
This over specification increases the representation space and increases search time
4
Issues with the GP Model
Search is not completely biased Mutation is always random Crossover is random given the current population
(driven by the current population but random there) Selection is the only bias
Selection drives the population and thus affects crossover
How to improve the bias? Building “models” and using them to drive the search
5
Constrained Genetic Programming (CGP)
An Option to Reduce Complexity
Developed by Dr. Cezary Janikow in 1996It uses user-specified “model” 1st order heuristics parent-child
Global at the root Local anywhere
Demonstrated to dramatically improve search time and complexity for problems such as SantaFe and 11-multiplexer
But the user has to give the “right” heuristics – discovery of the right heuristics can be a long process
6
Adaptable Constrained GP (ACGP)Automating the CGP Methodology
Developed by Dr. Cezary Janikow in 2003ACGP discovers 1st order heuristics Assumes that best trees are the “correct” heuristics Adjusts the heuristics at generational intervals Regrows the population using the new heuristics Mutation and crossover are now biased using the
heuristics Uses gradual adjustments of the heuristics to control
convergence
7
GP Building BlocksA Digression
8
The building block hypothesis states:Genetic processes work by combining relatively fit, short schema to form complete solutions.
a
b
Adaptable Constrained GP (ACGP)1st Order Heuristics
Controls one argument to a functionDevelops a set of heuristicsLooks at parent node (function) and one
child node (function or terminal)Treats those schema that are frequent in
the fittest population members as desirable
Gradual heuristic adjustment prevents premature convergence
9
Adaptable Constrained GP (ACGP)2nd Order Heuristics
Controls all arguments to a functionDevelops a set of heuristicsLooks at parent node (function) and all
child nodes (function or terminal)Treats those schema that are frequent in
the fittest population members as desirable
Gradual heuristic adjustment prevents premature convergence
10
Adaptable Constrained GP (ACGP)2nd Order Heuristics
Consider x*x 1st order can only speak of * and one argument at a
time Suppose that x* is 10% and that *x is 10% In the absence of heuristics about the entire subtree
x*x, the probability of x*x will be joint as 1% But the true probability of x*x in the problem can be
up to 10% Processing 1st order will always process it as 1% We need 2nd order (subtree) to process this as 10%
We need Mechanism to discover 1st and 2nd order structures Validate performance gains of 2nd order vs. 1st order
11
Where is the Advantage in ACGP?
Discovery of fit building blocks
Promotes the use of fit building blocks in crossover, mutation and regrow
Creates a population of highly fit compact solutions
ACGP discovers heuristics that improve fitness primarily but also reduce tree sizes
12
How ACGP Develops Its Heuristics
Tabulate the frequency of building blocks in the population
Tabulate the frequency of building blocks in the fittest individuals in the population
Increase the probability of fit heuristicsSuppress the probability of less fit
heuristicsAccess to heuristics is constant because of
the tabulation Disregarding paging
13
Representation of Heuristics in ACGP1st Order Heuristics
The general form of the 1st order combinations is
Using 4 binary functions, 14 terminals, this creates 162 combinations
14
Representation of Heuristics in ACGP2nd Order Heuristics
The general form of the 2nd order combinations is
Using 4 binary functions, 14 terminals, this creates 2592 combinations
15
Representation of Heuristics in ACGPHigher Order Heuristics
The general form of the 3rd order combinations is
Using 4 binary functions, 14 terminals, this creates 13,728,800 combinations
4th OH 3.76 * 1014 combinations 5th OH 2.84 * 1029 combinations 6th OH 1.62 * 1059 combinations
16
ACGP 2nd order Implementation was painfully validated
But experiments with Santafe and 11-multiplexer using 1st order and 2nd order gave the same results
Means that these problems do not have 2nd order structure not implied by joint probabilities
For validation, needed a function with obvious difference between 1st and 2nd order probabilities
17
Experimental Setup Target Equation: x*x + y*y + z*z
Has very explicit 2nd order structure different from that implied by 1st order
Function set: {+, -, *, /} (protected divide) Terminal set: {x, y, z, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5} Population size: 500 Generations: 500 Operators: crossover 85%, mutation 10%, selection 5%,
regrow 100% at each iteration Number of independent runs: 30 Fitness: sum of square errors on 100 random data points
in the range -10 to 10 Iteration length: 20 generations
18
ACGP Learning Results
19
Find the equation
Comparison of GP-Base, ACGP with 1st order and 2nd order heuristics.
Training slope-off with 50% update
ACGP Learning Results
20
Find the equation
Comparison of GP-Base, ACGP with 1st order and 2nd order heuristics.
Training slope-on
ACGP Learning Results
21
Find the equation
Comparison of GP-Base, ACGP with 1st order and 2nd order heuristics.
Training slope-on, first five iterations
ACGP Learning Results
22
Comparison of GP-Base, Strong 1st Order and Strong 2nd Order Heuristics on 200 generation, no iterations
Find the equation
ACGP Learning Results
23
Find the equation
Comparison of Base, 1st Order and 2nd Order Heuristics on a Time Scale
ACGP Learning Results
24
Average
Best Tree Size
Best Tree Depth
Execution Time
Base 728.40 19.43 347.6
1st OH 123.67 10.37 44.70
2nd OH 123.87 11.40 65.67
Find the equation
Average tree structure for Base, 1st Order and 2nd Order Heuristics
ACGP Learning Limits
25
Find the equation
Comparison of Base, 1st Order and 2nd Order Heuristics with Training Slope on for an extended Bowl3 equation
ACGP Learning ResultsDesirable 1st order heuristics
Implied 2nd order heuristics
Desirable 2nd order heuristics
26
{*1 , x} {*2 , x}{*1 , z} {*2 , z}{*2 , y}{*1 , y}{+1 , +} {+2 , +} {+2 , *}{+1 , *}
{*, x, x} {*, x, z} {*, y, x} {*, y, z}{*, x, y} {*, y, y}
{*, z, x} {*, z, z}{*, z, y}
{+, +, +} {+, *, +} {+, *, *}{+, +, *}
{*, x, x} {*, y, y} {*, z, z}{+, *, +} {+, *, *}{+, +, *}
ACGP Learning Results
27
1st order heuristics
2nd order heuristics
{*1 , x} {*2 , x}{*1 , z} {*2 , z}{*2 , y}{*1 , y}
0.2289 0.2426 0.2403 0.2323 0.2489 0.2087ACGP 1
{*, x, x} {*, x, z} {*, y, y} {*, z, z}{*, x, y} {*, y, z}
0.0532 0.0570 0.0478 0.0604 0.0506 0.0502ImpliedACGP 1
0.2545 0.0001 0.0001 0.2368 0.0001 0.2436ActualACGP 2
ACGP Learning Results
28
1st Order HeuristicsHeuristic Initial Final
‘*’
Left argX 0.056 0.2289Y 0.056 0.2426Z 0.056 0.2403
Right argX 0.056 0.2323Y 0.056 0.2489Z 0.056 0.2087
‘+’Left arg
‘*’ 0.056 0.4796‘+’ 0.056 0.2171
Right arg‘*’ 0.056 0.4168‘+’ 0.056 0.2410
Root ‘+’ 0.056 0.7669Average of all other heuristics 0.056 0.0371
First-order heuristics discovered. Root’s heuristics are zero-order for
Bowl3 equation
ACGP Learning Results
29
Second-order heuristics summary for ‘*’ for Bowl3 equation
Multiply Heuristics
Heuristic Initial Final
‘*’
X X 0.0031 0.2545
Y Y 0.0031 0.2368
Z Z 0.0031 0.2436
Average of all other heuristics
0.0031 0.0008
ACGP Learning Results
30
Second-order heuristics summary for ‘+’ for Bowl3 equation
Addition HeuristicsHeuristic Initial Final‘+’ ‘*’ ‘*’ 0.0031 0.3110‘*’ ‘+’ 0.0031 0.0688‘+’ ‘*’ 0.0031 0.1289Average of all other heuristics 0.0031 0.0015
Conclusions and Further Work
31
The 1st and 2nd order heuristics of ACGP provide a significant performance improvement over the base GP
If very strong second-order heuristics are present, ACGP is able to process them and also to discover them
Running 2nd order doesn’t hurt if there is no explicit 2nd order not implied by 1st
Overhead is minimal If a problem does not have explicit second-order
structure, ACGP running in first or second-order mode is the same
Going higher than 2nd order has drawbacks.The next step will be to move on to standard test
or real world problems
SummaryThe Genetic Programming (GP) ModelIssues with the GP ModelConstrained Genetic Programming (CGP)Adaptable Constrained GP (ACGP)Adaptable Constrained GP (ACGP)
HeuristicsACGP Learning ResultsConclusions
32
References Banzhaf, W., Nordin, P., Keller, R., Francone, F., Genetic Programming – An
Introduction, Morgan Kaufmann, 1998
Janikow, Cezary Z. A Methodology for Processing Problem Constraints in Genetic Programming, Computers and Mathematics with Applications. 32(8):97-113, 1996.
Janikow, Cezary Z., Deshpande, Rahul, Adaptation of Representation in GP. AMS 20032
Janikow, Cezary Z. ACGP: Adaptable Constrained Genetic Programming. In O’Reilly, Una-May, Yu, Tina, and Riolo, Rick L., editors. Genetic Programming Theory and Practice (II). Springer,
New York, NY, 2005, 191-206
Janikow, Cezary Z., and Mann, Christopher J. CGP Visits the Santa Fe Trail – Effects of Heuristics on GP. GECCO’05, June 25-29, 2005
Koza, John R. Genetic Programming. The MIT Press. 1992.
Koza, John R. Genetic Programming II. The MIT Press. 1994.
Looks, Moshe, Competent Program Evolution, Sever Institute of Washington University, December 2006 33
References Looks, Moshe, Competent Program Evolution, Sever Institute of Washington University,
December 2006
McKay, Robert I., Hoai, Nguyen X., Whigham, Peter A., Shan, Yin, O’Neill, Michael, Grammar-based Genetic Programming: a survey, Genetic Programming and Evolvable Machines, Springer Science + Business Media, September 2010
Poli, Riccardo, Langdon, William, B., Schema Theory for Genetic Programming with One-point Crossover and Point Mutation, Evolutionary Computation, MIT Press, Fall 1998
Sastry, Kumara, O’Reilly, Una-May, Goldberg, David, Hill, David, Building-Block Supply in Genetic Programming, IlliGAL Report No. 2003012, April 2003
Shan, Yin, McKay, Robert, Essam, Daryl, Abbass, Hussein, A Survey of Probabilistic Model Building Genetic Programming, The Artificial Life and Adaptive Robotics Laboratory, School of Information Technology and Electrical Engineering, University of New South Wales, Australia, 2005
Burke, Edmund, Gustafson, Steven, and Kendall, Graham. A survey and analysis of diversity measures in genetic programming. In Langdon, W., Cantu-Paz, E. Mathias, K., Roy, R., Davis, D., Poli, R., Balakrishnan, K., Honavar, V., Rudolph, G., Wegener, J., Bull, L., Potter, M., Schultz, A., Miller, J., Burke, E. and Jonoska, N., editors. GECCO2002: Proceedings of the Genetic and Evolutionary Computation Conference, 716-723, New York. Morgan Kaufmann.
34
References Daida, Jason, Hills, Adam, Ward, David, and Long, Stephen. Visualizing tree
structures in genetic programming. In Cantu-Paz, E., Foster, J., Deb, K., Davis, D., Roy, R., O’Reilly, U., Beyer, H., Standish, R., Kendall, G., Wilson, S., Harman, M., Wegener, J., Dasgupta, D., Potter, M., Schultz, A., Dowsland, K., Jonoska, N., and Miller, J., editors, Genetic and Evolutionary Computation – GECCO-2003, volume 2724 of LNCS, 1652-1664, Chicago. Springer Verlag.
Hall, John M. and Soule, Terence. Does Genetic Programming Inherently Adopt Structured Design Techniques? In O’Reilly, Una-May, Yu, Tina, and Riolo, Rick L., editors. Genetic Programming Theory and Practice (II). Springer, New York, NY, 2005, 159-174.
Langon, William. Quadratic bloat in genetic programming. In Whitley, D., Goldberg, D., Cantu-Paz, E., Spector, L., Parmee, I., and Beyer, H-G., editors, Proceedings of the Genetic and Evolutionary Conference GECCO 2000, 451-458, Las Vegas. Morgan Kaufmann.
McPhee, Nicholas F. and Hopper, Nicholas J. Analysis of genetic diversity through population history. In Banzhaf, W., Daida, J., Eiben, A. Garzon, M. Honavar, V., Jakiela, M. and Smith, R., editors Proceedings of the Genetic and Evolutionary Computation Conference, volume 2, pages 1112-1120, Orlando, Florida, USA. Morgan Kaufmann.
Franz Rothlauf. Representations for Genetic and Evolutionary Algorithm. Springer, 2010.
35