笔者想入门运动科学/运动分析这个领域,打算从阅读一篇综述开始分析这个领域的问题、方法。
题目: Meta-heuristics meet sports
a systematic review from the viewpoint of nature inspired algorithms
本篇为第一part,包含了introduction和一些evolutionary algorithms。
introduction:
optimization
例子:player selection, optimum player lineup, player ranking / finding player roles
难点:效率
传统方法:
interior point method
trust region method
问题
1.局部最优
2.初始情况
3.微分计算
一种解决方法:
meta-heuristics
特点stochastic, nature inspired)
分类 EA(evolutionary algorithms), SI(swarm intelligence), PI(physically inspired algorithms) …
EA
GA(genetic) Hayes-Roth, 1975
DE(differential evolution) Storn and Price, 1997
SI
近十几年
PSO(particle swarm optimization) Kennedy and Eberhart, 1995
ACO(ant colony optimization) Dorigo and Gambardella, 1997
FA(firefly algorithm) Yang, 2009
CS(cuckoo search) Yang and Deb, 2009
GWO(grey wolf optimization) Mirjalili et al. 2014
怎么处理搜索空间问题:
多个相互影响的agent(multiple interacting agent)
简单又高效。处理了实际的NP问题
Marano et al., 2011
Yang et al., 2013
Meta-heuristics在sport领域应用
player selection, battling line ups, categorizations
hybrid versions Nelikanti et al 2021
variants of the meta-heuristics Manafifard et al 2015
Therefore the aim of this paper is twofold: …, and …
PRISMA(prefered Reporting Items for Systematic Reviews and Meta- Analyses) guidelines Moher et Al. 2009
keyword
●sports& meta-heuristics,
●genetic algorithm, firefly algorithm, Ant colony system, particle swarm optimization,
●cricket, soccer, basketball…
●hybrid
research work repositorires
●google scholar
●scopus
●semantic scholar
搜索过程
1.时间
a.2021.8~2022.4 <= there are limited research done on selected topic
b.publications > 1970 <= 这些仿生算法大部分在1970年之后才被提出
2.语言 不限 但是找到的大部分是英语
3.方向
a.team selection, optimal lineups, sports equipment, optimization, scheduling and ranking, performance analysis, predictions in sports, play tracking
b.sports gambling optimization, sports tournaments optimization are excluded, out of the scope of this work
4.工具 Rayyan:一个流行的web application, dedicated to systematic reviews(Ouzzani et al. 2016). Allows fast article navigation and summarization of the relevant articles. They used Rayyan to organize and extract relevant facts.
Rayyan (Ouzzani et al. 2016)
5.分类:按照具体的优化任务,而不是使用的meta-heuristic算法。 因为本文目标是找到仿生优化算法对运动领域的贡献以及这种贡献是怎么做到的。
6.此外,还有以下几个方面:
a.算法怎么实现的(implement)
b.问题怎么编码的(representation, encode)
c.适应度评估方法是什么(fitness evaluation)
d.算法特定的算子用的是什么(operator)
然后采用PRISMA的comprehensive review process进行评估
meta-heuristic特征
●swarm behaviour
●survival of the fitness
●population based algorithms
●iteratively in stochastic nature while applying different operators that mimic different natural optimization processes
●balanced use of exploration and exploitation properties, essential for finding promising solutions. exploration-> global search; whereas exploitation -> local search
来源:meta-heuristic: a higher level procedures or heuristics which are capable of providing sufficient good solutions to optimization problems.
EA
variants that are better adapted to their environments are more likely to survive and reproduce
●mutation & crossover: exploration
●selection: exploitation
GA
most significant and widely used
chromosome/variant/a feasible solution to problem
representation: binary, real value, integer, permutation
DE
optimize real parameter, real value functions aloowing continuous optimization
3 main types of vectors: donor vector, target vector, trial vector
Start: a random initial vectors supposed to be feasible solutions
target vector <= a selected population member in that instance
Mutation: for a particular target vector, DE generates a donor vector by adding the weighted difference between two population vectors to a third vector where all three are different and distinct from the target vector as well
donor vector: represents the direction and magnitude of the change caused by a mutation
Crossover: a trial vector is then generated using the target vector and the donor vectorThe survival selection leads finding the best among the target vector(i.e. parent) and the trial vector( i.e. offspring created from the target vector and donor vector) based on the fitness value
Similar to the survivor selection mechanism in PSO where the offspring must be equal to or outperform the corresponding parent to survive
MA
memetic algorithms
use additional local search technique to reduce the likelyhood of the premature convergence
名字与meme有关
mimic a cultural evolution which is accomplished in MAs by incorporating heuristics, approximation algorithms, local search techniques etc.
The inclusion of problem domain knowledge can be mentioned as the most crucial and the distinctive feature of MAs
Burke et al., 1995, Knowles and Corne, 2000, Lu¨ and Hao, 2010
ES
最早的处理高维连续值域问题的算法之一
只有一个突变变量
ES uses only a mutation parameter. 只有一个parameter用于突变还是只有一个突变,没有crossover?
(1+1) ES (pronounced as one plus one ES): 最早的ES,simple two-membered algorithms
offspring
GP
为了解决如何让计算机能不靠人工干预地解决问题
GP searches the space of possible computer programs for a program that is highly fit for solving the problem at hand.
很多使用LISP语言,应为他能用(操纵某种)data structure操纵source code
通常会选择出来small program functions within a larger application
Espejo et al., 2009
Langdon and Harman, 2014
Weimer et al., 2009
累了。下一节从swarm intelligence algorithms开始。