1. introduction--Evolutionary algorithm

This series of blogs begins by introducing evolutionary algorithms.

The content of knowledge comes from the course I listen to in my Master's degree and the literature books I read. It's a lecture note or a reading note.

The books that bloggers read abroad may be mixed in Chinese and English.

Outline

1. What is evolutionary algorithm
2. What kind of problems can be solved
3. An important part of evolutionary algorithms
4. Eight Queens question (example)

1. What are evolutionary algorithms

Genetic Algorithm (GA) is a computational model for simulating the evolutionary process of organisms, and is a new computational method for the combination of natural genetics and computer science.

< pictures from, Frank Neumann, the University of adelaide>

2. What kind of problems can be solved

The three main types of problems we face:

2.1 Optimisation

We have a system model, but we need to find the right input to achieve the goal we want.

2.2 Modelling

We already have a lot of input and output, and now we need to find a suitable model for each input to get the correct output.

2.3 Simulation

We already have a given model and now we want to get the output results in different input cases.

3. An important component of evolutionary algorithms (components of evolutionary algorithms)

• Expression
• Evaluation algorithm/Adaptation algorithm
• Population
• Parental selection mechanism
• Variation operators
• Supervisory selection mechanism
• Initialization
• Termination conditions

Some things are more abstract, and later chapters are detailed.

3.1 Expression (representation)

Phenotype and genotype (phenotypes and genotypes)

Directly for example:

For example, in an optimization problem, all possible solutions are integer (int), so a given integer is a set of phenotype.

In this example, these integers can be represented by binary, so for example 18 is a phenotype, then 10010 is genotype.

< so for example, but not that genotype is binary >

Genotype (genotype)--by decoding (decode)----phenotype (phenotype)

Our goal is to get the ultimate best solution, so this solution is obtained by decode the best genotype.

 

3.2 Evaluation algorithm/adaptation algorithm (Evaluation function/fitness function)

The role of the evaluation function is to represent the requirements, the population should adapt to.

Simply put, this evolutionary algorithm is used to solve the problem of how to improvement.

3.3 Population (Population)

A population is a multiset of genotypes.

Population can tolerate all possible solution, including many genotype sets.

Individual (individuals), the equivalent of static objects, can not change or adapt, is to rely on population to adapt and change.

3.4 Parental selection mechanism (parent selection mechanism)

Parental selection mechanism is used to distinguish a good quality from a large number of individuals.

The better individuals, who transcend their parents ' individual, will be used as new parents to generate new generations.

3.5 Variation Operators

The role of variation operators is to create new individuals from the old individuals. The new candidate solution is created according to the phenotype.

This piece contains mutations (mutation) and recombination (recombination)

The mutation is about a genotype that produces a variant (or descendants) that accompanies a series of random choices.

This can be seen as a blood supply for gene pool (fresh blood).

Variation can be understood as a variation of a dollar operator

The reorganization is about mixing the parents ' genotypes into the genotypes of future generations. As with mutations, the parts that merge and merge are also random.

The principle behind recombination is simple-by mating and individuals with different but desirable features, we can pro Duce an offspring that combines both of those features.

The reorganization can be understood as two Yuan variation operator

3.6 Supervisory Selection mechanism (Survivor selection mechanism)

The function of the supervisory selection mechanism is to distinguish the better part from the many individuals.

and Parents choose the difference:

1. The stage of application is different. The supervisory selection mechanism will be used when new generations are created.

2. Parental selection is usually random (stochastic), while the supervisory selection mechanism is deterministic (deterministic)

3.7 Initialization (initialisation)

So much evolution, how did the first generation of population come about?

The first population is seeded by randomly generated individuals.

This is quite magical, it is so random out of ... We'll talk about it later.

3.8 Termination conditions (termination condition)

has been evolving, that must have an end, like recursion, there should be a termination condition.

In some of the problems, the problem will give a known standard, the known optimization criteria, when reached or to a certain extent close to the standard, we can stop.

However, in the EA algorithm often has a lot of randomness, so there is no guarantee that the standard can be achieved ... It seems that the operation will never stop. Therefore, in order to solve this problem, we also provide the following criteria as the criteria for stopping the algorithm:

• The maximally allowed CPU time elapses.
• The total number of fitness evaluations reaches a given limit.
• The fitness improvement remains under a threshold value for a given period of time (i.e., for a number of generations or F Itness evaluations).
• The population diversity drops under a given threshold.

The concept is more, with a picture summary:

Pseudo code:

4. Eight Queen's question

The above is too abstract, here is a simple example.

What is the eight Queens question? See Baidu Encyclopedia, here to introduce algorithms.

4.1 Expression:

From the picture, the phenotype is the situation shown on the board. The genotype is the set of numbers 13526478 (each number represents the position on the board)

4.2 Variants

4.3 reorganization

4.4 Parental selection and supervision options

4.5 Summary

References: main source notes and professors ' ppt. Frank Neumann, the University of Adelaide

1. introduction--Evolutionary algorithm