An Introduction to Genetic Algorithm (GA) By: Dola Pathak For: STAT 992:Computational Statistics SPRING

Motivating Example The data were collected at Baystate Medical Center in Spring- field, Massachusetts, in 1986, on 189 births and the response variable was the birth weight of the baby. There were eight explanatory variables/factors which were considered. This data, called the low birth weight data, is in the Appendix I of Hosmer and Lemeshow’s book “Applied Logistic Regression”. What is the best fitted model? Are all the variables important? How many models are possible with main effects only – 256! Any alternatives? 2

What is Genetic Algorithm (GA)? A genetic algorithm is an optimization method based on natural selection. 3

The concept Inspired by Darwin’s Theory of evolution popularly termed as “Survival of the fittest” Fittest – ability to adapt to the environment which increases the likelihood of survival The fitness is maintained through the process of reproduction, crossover and mutation. 4

From the statistical point of view A GA is a stochastic technique with simple operations based on the theory of natural selection. The basic operations are : 1. Selecting population members for the next generation 2. Mating these members via crossover of “chromosomes” In statistical terms chromosomes will be the individual members of the population and the genes of the chromosomes are the variables. 3. Performing mutations on the chromosomes to preserve population diversity so as to avoid convergence to local optima. 4. Finally, determining the fitness of each member in the new generation using an evaluation (fitness) function. This fitness influences the selection process for the next generation. 5

Development Alan Turing -----“learning machine” evolution process In 1960s the idea of evolution strategy/programming was developed in Germany and in the USA. John Holland and his students in the early 1970s (book: "Adaption in Natural and Artificial Systems"(1975)). David Goldberg, who was a graduate student of Holland’s presented an award-winning doctoral thesis on his GA application to gas pipeline optimization. 6

Why Genetic Algorithm? Exhaustive search (where every possible combination is checked ) - accurate result - time consuming - inefficient when the numbers of variables are large - optimizers tend to : - limit the number of variables they use or - limit the number of values these variables can take. The genetic algorithm (does not try every possible combination ) - optimal solution by selecting variables intelligently. - far more variables can be utilized - all values of a variable can be checked - simultaneous searching - less likely to become stuck in "local minima" - less time - efficient 7

Where are Genetic Algorithms applied? There are at least three situations where genetic algorithms are applied: 1. The objective function is not smooth (i.e., not differentiable). 2. There are multiple local optima( often happens with non linear functions). 3. There are a large number of parameters (the meaning of “large” keeps changing). 8

Outline of the Basic Genetic Algorithm [Start] Generate random population of n chromosomes (suitable solutions for the problem) [Fitness] Evaluate the fitness f(x) of each chromosome x in the population [New population] Create a new population by repeating the following steps until the new population is complete [Selection] Select two parent chromosomes from a population according to their fitness (the better fitness, the bigger chance to be selected) 9

[Crossover] With a crossover probability cross over the parents to form a new offspring (children). If no crossover was performed, offspring is an exact copy of parents. [Mutation] With a mutation probability mutate new offspring at each locus (position in chromosome). [Accepting] Place new offspring in a new population [Replace] Use new generated population for a further run of algorithm [Test] If the end condition is satisfied, stop, and return the best solution in current population [Loop] Go to step 2 10

11 Scrucca (April, 2013)

Example 1 Going back to the example that was presented at the beginning of the presentation, let us see how a first generation of optimum variable values are calculated using a Genetic Algorithm. 12

13

STEP1: Generate random population of 4 chromosomes (individuals) > set.seed(7777) > x y z population_chrom<-matrix(c(x,y,z),byrow = TRUE,nrow=3,ncol=4) #population units PARENT POPULATION population_chrom [,1] [,2] [,3] [,4] [1,] [2,] [3,] STEP2 Evaluate the fitness f(x) of each chromosome x in the population > fitness<-fit(population_chrom) > fitness [,1] [,2] [,3] [,4] [1,]

STEP 3: Select two parent chromosomes from a population according to their fitness (the better fitness, the bigger chance to be selected) Determining the new ordering > prob<-objective/total #probabilities associated with each chromosome > set.seed(9001) > gen_prob<-runif(4, min=0, max=1) > data.frame(t(prob), gen_prob) t.prob. gen_prob #generating the new order of the parent population (chromosomes) for the 1st generation > Chrom_reordered [,1] [,2] [,3] [,4] [1,] [2,] [3,]

16

Crossover: After Crossover: > Chromosome_cross [,1] [,2] [,3] [,4] [1,] [2,] [3,] > Chromosome_cross [,1] [,2] [,3] [,4] [1,] [2,] [3,]

18

1 st Generation: > first_gen<-Chromosome_cross > first_gen [,1] [,2] [,3] [,4] [1,] [2,] [3,] Comparing the fitness function: > fitness_new<-fit(first_gen) > data.frame(t(fitness),t(fitness_new)) t.fitness. t.fitness_new

20

CASE 2: GA Search Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) < 2e-16 *** race.catBlack ** race.catOther ** preterm lwt * sm ** ht ** ui *** COMPARISON: 21

With pairwise interaction: CASE 1:Exhaustive Search For the Exhaustive case the iterations went on for a long time and it failed to give the best fitted model. The program ran for approximately 710 minutes before it was terminated. 22

CASE 2: GA Search Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) e-13 *** race.catBlack * race.catOther preterm age ** lwt sm age:sm * age:ht * lwt:ht age:ui e-05 *** preterm1+:lwt age:ftv.catNone ** age:ftv.catMany * lwt:ftv.catNone ** lwt:ftv.catMany Signif. codes: 0 ‘***’ ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 23

Comparison of time taken for the methods to converge: Conclusion: It is seen that for models with pairwise interactions, GA gets to the best fitted model much faster than the exhaustive search. MethodExhaustive SearchGenetic Algorithm Time taken to find the best fit model minutes (failed to give the optimum solution) minutes (260 generations) 24

GAs are used for Building Double Threshold Auto Regressive Conditional Heteroscedastic (DTARCH) models Multilayer neural network Optimal Component Selection for component based systems Robotics VLSI layout optimization Modelling the behavior of economic agents in macroeconomic models, etc. 25

Computers and GA: With the advent of different packages and softwares the use of GA has definitely gained popularity. Apart from glmulti, some of the packages which I came across while researching for this presentation were:  GA  genalg  PROC GA  PROC CONNECT ( uses Parallel Genetic Algorithm)  Genetic Algorithm Toolbox in MATLAB  GAlib (available for the UNIX system) 26

References and further readings: Denny Hermawanto. Genetic Algorithm for Solving Simple Mathematical Equality Problem. Hillier and Lieberman. Introduction to Operations Research (9 th Ed.) Hosmer, D.W. and Lemeshow, S. Applied Logistic Regression. New York: Wiley Givens & Hoeting.Computational Statistics (2 nd Ed.) K.F Man, et all. Genetic Algorithms: Concepts and Applications. IEEE Transactions on Industrial Electronics, Vol.43 No 5, Oct 1996 Luca Scrucca. GA: A Package for Genetic Algorithms in R. Journal of Statistical Software, Vol 23, Issue 4, April 2013 V. Calcagno, et all. glmulti: An R Package for Easy Automated Model Selection with (Generalized) Linear Models. Journal of Statistical Software, Vol 24 Issue 12, May / 27

Thank you 28