Picking a cricket 11 via genetic algorithm

‘Cricket team selection using Genetic Algorithm’. This is the title of a paper accepted by the International Scientific Committee of the International Congress on Sports Dynamics held in Melbourne on September 1. The paper was presented by S N Omkar, the yoga guru of the Indian cricket team.

Omkar believes his procedure can help pick a team of 15 or so from the hundreds of aspirants. The theory, formulated along with R Verma of Regional Engineering College, Warangal, takes into account performance and fitness data of players and gives an optimised and balanced team with right number of batsmen, bowlers and wicket-keepers without any human interference.

The other paper submitted by Omkar dwells on ‘Yogic Sun salutation for better cricketing’. The Bangalore-based IISc scientist explains his work in an exclusive interview to The Times of India . Excerpts:

Could you explain in simple terms the process of team selection through genetic algorithm?

Genetic algorithms are inspired by Darwin’s theory of evolution. Simply said, problems are solved by an evolutionary process resulting in a best (fittest) solution (survivor). In other words, the solution is evolved.

The fitness of an organism is measured by its success in its life (survival). Algorithm begins with a set of solutions — represented by chromosomes — called population. In the context of the current problem, each population would consist of team cricketers randomly picked from the probables list. Several such teams, say 2000, are formed. A fitness function is evaluated for each team. The fitness function is based on the performance, health condition, experience, etc., of every member in the team. Constraints like number of spinners or fast bowlers required, scope for new players etc., can also be included.

Based on the fitness of each team, new teams with better fitness values are formed by using some operations like crossover and mutation. The new generation of teams thus formed are most likely to be better than the previous ones. This process of forming new generation of teams is continued until a best fitness value is obtained.

How is this procedure different from an evaluation of batting, bowling fielding performances with the help of statistics? What makes it foolproof?

Genetic algorithms belong to the class of stochastic search methods. It optimises the fitness function chosen, thus differing from statistics. Genetic algorithms operate on a population of solutions and in many practical problems they seem to yield good results.

Take for instance a batsman A (somebody like Srikkanth), who is flashy and wants to strike every ball, and musters up an average of 40. And another batsman B (somebody like Shiv Sundar Das perhaps) who is cautious and judicious, but has a batting average of 35. Let’s assume other factors are equal. In this case, how does the method work? Who would gain entry into the team?

This can be taken care while defining the fitness or objective function. For example, objective function can include a parameter (strike rate) and more weightage can be given in case one is selecting the team for one-dayers.

Apart from this one can also include factors like performance of any individual against a particular team (within the country /abroad) in the past. Also, an individual’s performance on a certain pitch can also be included in the fitness function.

Does this method take into consideration rival strengths too? This is vital in a ‘human’ selection procedure. For example, if Australia has five left-handers, would we go in for two off-spinners? Again, what kind of spinners __ finger spinners or tweakers?

It is possible. All these depend on how one defines the fitness or objective function.

Can this method evaluate who’s the better bat — Sachin Tendulkar and Brian Lara?

This method is used for optimization.


Comments are closed.

%d bloggers like this: