An article has been making the rounds lately discussing the use of genetic algorithms to optimize "build orders" in StarCraft II.
http://lbrandy.com/blog/2010/11/using-genetic-algorithms-to-find-starcraft-2-build-orders/
The initial state of a StarCraft match is pre-determined and constant. And like chess, decisions made in this early stage of the match have long-standing consequences to a player's ability to perform in the mid and late game. So the various opening possibilities or "build orders" are under heavy study and scrutiny. Until the circulation of the above article, computer-assisted build order creation probably wasn't as popularity as it has been recently.
My question is... Is a genetic algorithm really the best way to model optimizing build orders?
A build order is a sequence of actions. Some actions have prerequisites like, "You need building B before you can create building C, but you can have building A at any time." So a chromosome may look like AABAC.
I'm wondering if a genetic algorithm really is the best way to tackle this problem. Although I'm not too familiar with the field, I'm having a difficult time shoe-horning the concept of genes into a data structure that is a sequence of actions. These aren't independent choices that can be mixed and matched like a head and a foot. So what value is there to things like reproduction and crossing?
I'm thinking whatever chess AIs use would be more appropriate since the array of choices at any given time could be viewed as tree-like in a way.
Although I'm not too familiar with the field, I'm having a difficult time shoe-horning the concept of genes into a data structure that is a sequence of actions. These aren't independent choices that can be mixed and matched like a head and a foot. So what value is there to things like reproduction and crossing?
Hmm, that's a very good question. Perhaps the first few moves in Starcraft can indeed be performed in pretty much any order, since contact with the enemy is not as immediate as it can be in Chess, and therefore it is not as important to remember the order of the first few moves as it is to know which of the many moves are included in those first few. But the link seems to imply otherwise, which means the 'genes' are indeed not all that amenable to being swapped around, unless there's something cunning in the encoding that I'm missing.
On the whole, and looking at the link you supplied, I'd say that genetic algorithms are a poor choice for this situation, which could be accurately mathematically modelled in some parts and the search tree expanded out in others. They may well be better than an exhaustive search of the possibility space, but may not be - especially given that there are multiple populations and poorer ones are just wasting processing time.
However, what I mean by "a poor choice" is that it is inefficient relative to a more appropriate approach; that's not to say that it couldn't still produce 98% optimal results in under a second or whatever. In situations such as this where the brute force of the computer is useful, it is usually more important that you have modelled the search space correctly than to have used the most effective algorithm.
As TaslemGuy pointed out, Genetic Algorithms aren't guaranteed to be optimal, even though they usually give good results.
To get optimal results you would have to search through every possible combination of actions until you find the optimal path through the tree-like representation. However, doing this for StarCraft is difficult, since there are so many different paths to reach a goal. In chess you move a pawn from e2 to e4 and then the opponent moves. In StarCraft you can move a unit at instant x or x+1 or x+10 or ...
A chess engine can look at many different aspects of the board (e.g. how many pieces does it have and how many does the opponent have), to guide it's search. It can ignore most of the actions available if it knows that they are strictly worse than others.
For a build-order creator only time really matters. Is it better to build another drone to get minerals faster, or is it faster to start that spawning pool right away? Not as straightforward as with chess.
These kinds of decisions happen pretty early on, so you will have to search each alternative to conclusion before you can decide on the better one, which will take a long time.
If I were to write a build-order optimizer myself, I would probably try to formulate a heuristic that estimates how good (close the to the goal state) the current state is, just as chess engines do:
Score = a*(Buildings_and_units_done/Buildings_and_units_required) - b*Time_elapsed - c*Minerals - d*Gas + e*Drone_count - f*Supply_left
This tries to keep the score tied to the completion percentage as well as StarCraft common knowledge (keep your ressources low, build drones, don't build more supply than you need). The variables a to f would need tweaking, of course.
After you've got a heuristic that can somewhat estimate the worth of a situation, I would use Best-first search or maybe IDDFS to search through the tree of possibilities.
Edit:
I recently found a paper that actually describes build order optimization in StarCraft, in real time even. The authors use depth-first search with branch and bound and heuristics that estimate the minimum amount of effort required to reach the goal based on the tech tree (e.g. zerglings need a spawning pool) and the time needed to gather the required minerals.
Genetic Algorithm can be, or can sometimes not be, the optimal or non-optimal solution. Based on the complexity of the Genetic Algorithm, how much mutation there is, the forms of combinations, and how the chromosomes of the genetic algorithm is interpreted.
So, depending on how your AI is implemented, Genetic Algorithms can be the best.
You are looking at a SINGLE way to implement genetic algorithms, while forgetting about genetic programming, the use of math, higher-order functions, etc. Genetic algorithms can be EXTREMELY sophisticated, and by using clever combining systems for crossbreeding, extremely intelligent.
For instance, neural networks are optimized by genetic algorithms quite often.
Look up "Genetic Programming." It's similar, but uses tree-structures instead of lines of characters, which allows for more complex interactions that breed better. For more complex stuff, they typically work out better.
There's been some research done using hierarchical reinforcement learning to build a layered ordering of actions that efficiently maximizes a reward. I haven't found much code implementing the idea, but there are a few papers describing MAXQ-based algorithms that have been used to explicitly tackle real-time strategy game domains, such as this and this.
This Genetic algorithm only optimizes the strategy for one very specific part of the game: The order of the first few build actions of the game. And it has a very specific goal as well: To have as many roaches as quickly as possible.
The only aspects influencing this system seem to be (I'm no starcraft player):
build time of the various units and
buildings
allowed units and buildings given the available units and buildings
Larva regeneration rate.
This is a relatively limited, relatively well defined problem with a large search space. As such it is very well suited for genetic algorithms (and quite a few other optimization algorithm at that). A full gene is a specific set of build orders that ends in the 7th roach. From what I understand you can just "play" this specific gene to see how fast it finishes, so you have a very clear fitness test.
You also have a few nice constraints on the build order, so you can combine different genes slightly smarter than just randomly.
A genetic algorithm used in this way is a very good tool to find a more optimal build order for the first stage of a game of starcraft. Due to its random nature it is also good at finding a surprising strategy, which might have been an additional goal of the author.
To use a genetic algorithm as the algorithm in an RTS game you'd have to find a way to encode reactions to situations rather than just plain old build orders. This also involves correctly identifying situations which can be a difficult task in itself. Then you'd have to let these genes play thousands of games of starcraft, against each other and (possibly) against humans, selecting and combining winners (or longer-lasting losers). This is also a good application of genetic algorithms, but it involves solving quite a few very hard problems before you even get to the genetic algorithm part.
Related
I would like to calculate 3D balanced paths to building exits at once, for huge amount of people (2000). Since the problem is related to evacuation and solutions for 3D paths (the fastest and others) can be precalcualted and I am going to store 3D paths in database to accelerate the process. As I see it, there are two solutions so far:
Calculation of a number of passing through nodes, in graph environment representation, but probably the time calculation will be intolerable.
Using GA. However, I cannot find a good described optimization example, where is used genetic algorithm.
Can you tell me a way of using GA for multiobjective optimization, because I found only implementation of GA for finding shortest path? and Which algorithm is the best for multi-object optimization?
Genetic Algorithm as it is cannot be easily used for multi-objective optimisation directly. If you want to use the purest GA you have to combine the objectives into a single objective, e.g. by summing weighted values of each objective. But this often does not work very well, especially when there is a strong tradeoff between the objectives.
However, there are genetic (evolutionary) algorithms designed specifically for multi-objective optimisation. Probably the most famous one and one of the best is the NSGA-II which stands for Nondominated Sorting Genetic Algorithm II. It's also relatively easy to implement (I did once). But there other MOEAs (Multi-Objective Evolutionary Algorithm) too, just google it. Some of them also use the nondomination idea, others not.
When looking at Genetic programming papers, it seems to me that the number of test cases is always fixed. However, most mutations should (?) at every stage of the execution be very deleterious, i. e. make it obvious after one test case that the mutated program performs much worse than the previous one. What happens if you, at first, only try very few (one?) test case and look whether the mutation makes any sense?
Is it maybe so that different test cases test for different features of the solutions, and one mutation will probably improve only one of those features?
I don't know if I agree with your assumption that most mutations should be very deleterious, but you shouldn't care even if they were. Your goal is not to optimize the individuals, but to optimize the population. So trying to determine if a "mutation makes any sense" is exactly what genetic programming is supposed to do: i.e. eliminate mutations that "don't make sense." Your only "guidance" for the algorithm should come through the fitness function.
I'm also not sure what you mean with "test case", but for me it sounds like you are looking for something related to multi-objective-optimization (MOO). That means you try to optimize a solution regarding different aspects of the problem - therefore you do not need to mutate/evaluate a population for a specific test-case, but to find a multi objective fitness function.
"The main idea in MOO is the notion of Pareto dominance" (http://www.gp-field-guide.org.uk)
I think this is a good idea in theory but tricky to put into practice. I can't remember seeing this approach actually used before but I wouldn't be surprised if it has.
I presume your motivation for doing this is to improve the efficiency of the applying the fitness function - you can stop evaluation early and discard the individual (or set fitness to 0) if the tests look like they're going to be terrible.
One challenge is to decide how many test cases to apply; discarding an individual after one random test case is surely not a good idea as the test case could be a real outlier. Perhaps terminating evaluation after 50% of test cases if the fitness of the individual was <10% of the best would probably not discard any very good individuals; on the other hand it might not be worth it given a lot of individuals will be of middle-of-the road fitness and might well only save a small proportion of the computation. You could adjust the numbers so you save more effort, but the more effort you try to save the more chances you have of genuinely good individuals being discarded by accident.
Factor in the extra time to taken to code this and possible bugs etc. and I shouldn't think the benefit would be worthwhile (unless this is a research project in which case it might be interesting to try it and see).
I think it's a good idea. Fitness evaluation is the most computational intense process in GP, so estimating the fitness value of individuals in order to reduce the computational expense of actually calculating the fitness could be an important optimization.
Your idea is a form of fitness approximation, sometimes it's called lazy evaluation (try searching these words, there are some research papers).
There are also distinct but somewhat overlapping schemes, for instance:
Dynamic Subset Selection (Chris Gathercole, Peter Ross) is a method to select a small subset of the training data set on which to actually carry out the GP algorithm;
Segment-Based Genetic Programming (Nailah Al-Madi, Simone Ludwig) is a technique that reduces the execution time of GP by partitioning the dataset into segments and using the segments in the fitness evaluation process.
PS also in the Brood Recombination Crossover (Tackett) child programs are usually evaluated on a restricted number of test cases to speed up the crossover.
I'm looking for ideas/experiences/references/keywords regarding an adaptive-parameter-control of search algorithm parameters (online-learning) in combinatorial-optimization.
A bit more detail:
I have a framework, which is responsible for optimizing a hard combinatorial-optimization-problem. This is done with the help of some "small heuristics" which are used in an iterative manner (large-neighborhood-search; ruin-and-recreate-approach). Every algorithm of these "small heuristics" is taking some external parameters, which are controlling the heuristic-logic in some extent (at the moment: just random values; some kind of noise; diversify the search).
Now i want to have a control-framework for choosing these parameters in a convergence-improving way, as general as possible, so that later additions of new heuristics are possible without changing the parameter-control.
There are at least two general decisions to make:
A: Choose the algorithm-pair (one destroy- and one rebuild-algorithm) which is used in the next iteration.
B: Choose the random parameters of the algorithms.
The only feedback is an evaluation-function of the new-found-solution. That leads me to the topic of reinforcement-learning. Is that the right direction?
Not really a learning-like-behavior, but the simplistic ideas at the moment are:
A: A roulette-wheel-selection according to some performance-value collected during the iterations (near past is more valued than older ones).
So if heuristic 1 did find all the new global best solutions -> high probability of choosing this one.
B: No idea yet. Maybe it's possible to use some non-uniform random values in the range (0,1) and i'm collecting some momentum of the changes.
So if heuristic 1 last time used alpha = 0.3 and found no new best solution, then used 0.6 and found a new best solution -> there is a momentum towards 1
-> next random value is likely to be bigger than 0.3. Possible problems: oscillation!
Things to remark:
- The parameters needed for good convergence of one specific algorithm can change dramatically -> maybe more diversify-operations needed at the beginning, more intensify-operations needed at the end.
- There is a possibility of good synergistic-effects in a specific pair of destroy-/rebuild-algorithm (sometimes called: coupled neighborhoods). How would one recognize something like that? Is that still in the reinforcement-learning-area?
- The different algorithms are controlled by a different number of parameters (some taking 1, some taking 3).
Any ideas, experiences, references (papers), keywords (ml-topics)?
If there are ideas regarding the decision of (b) in a offline-learning-manner. Don't hesitate to mention that.
Thanks for all your input.
Sascha
You have a set of parameter variables which you use to control your set of algorithms. Selection of your algorithms is just another variable.
One approach you might like to consider is to evolve your 'parameter space' using a genetic algorithm. In short, GA uses an analogue of the processes of natural selection to successively breed ever better solutions.
You will need to develop an encoding scheme to represent your parameter space as a string, and then create a large population of candidate solutions as your starting generation. The genetic algorithm itself takes the fittest solutions in your set and then applies various genetic operators to them (mutation, reproduction etc.) to breed a better set which then become the next generation.
The most difficult part of this process is developing an appropriate fitness function: something to quantitatively measure the quality of a given parameter space. Your search problem may be too complex to measure for each candidate in the population, so you will need a proxy model function which might be as hard to develop as the ideal solution itself.
Without understanding more of what you've written it's hard to see whether this approach is viable or not. GA is usually well suited to multi-variable optimisation problems like this, but it's not a silver bullet. For a reference start with Wikipedia.
This sounds like hyper heuristics which you're trying to do. Try looking for that keyword.
In Drools Planner (open source, java) I have support for tabu search and simulated annealing out the box.
I haven't implemented the ruin-and-recreate-approach (yet), but that should be easy, although I am not expecting better results. Challenge: Prove me wrong and fork it and add it and beat me in the examples.
Hyper heuristics are on my TODO list.
Something pretty annoying in evolutionary computing is that mildly different and overlapping concepts tend to pick dramatically different names. My latest confusion because of this is that gene-expression-programming seems very similar to cartesian-genetic-programming.
(how) Are these fundamentally different concepts?
I've read that indirect encoding of GP instructions is an effective technique ( both GEP and CGP do that ). Has there been reached some sort of consensus that indirect encoding has outdated classic tree bases GP?
Well, it seems that there is some difference between gene expression programming (GEP) and cartesian genetic programming (CGP or what I view as classic genetic programming), but the difference might be more hyped up than it really ought to be. Please note that I have never used GEP, so all of my comments are based on my experience with CGP.
In CGP there is no distinction between genotype and a phenotype, in other words- if you're looking at the "genes" of a CGP you're also looking at their expression. There is no encoding here, i.e. the expression tree is the gene itself.
In GEP the genotype is expressed into a phenotype, so if you're looking at the genes you will not readily know what the expression is going to look like. The "inventor" of GP, Cândida Ferreira, has written a really good paper and there are some other resources which try to give a shorter overview of the whole concept.
Ferriera says that the benefits are "obvious," but I really don't see anything that would necessarily make GEP better than CGP. Apparently GEP is multigenic, which means that multiple genes are involved in the expression of a trait (i.e. an expression tree). In any case, the fitness is calculated on the expressed tree, so it doesn't seem like GEP is doing anything to increase the fitness. What the author claims is that GEP increases the speed at which the fitness is reached (i.e. in fewer generations), but frankly speaking you can see dramatic performance shifts from a CGP just by having a different selection algorithm, a different tournament structure, splitting the population into tribes, migrating specimens between tribes, including diversity into the fitness, etc.
Selection:
random
roulette wheel
top-n
take half
etc.
Tournament Frequency:
once per epoch
once per every data instance
once per generation.
Tournament Structure:
Take 3, kill 1 and replace it with the child of the other two.
Sort all individuals in the tournament by fitness, kill the lower half and replace it with the offspring of the upper half (where lower is worse fitness and upper is better fitness).
Randomly pick individuals from the tournament to mate and kill the excess individuals.
Tribes
A population can be split into tribes that evolve independently of each-other:
Migration- periodically, individual(s) from a tribe would be moved to another tribe
The tribes are logically separated so that they're like their own separate populations running in separate environments.
Diversity Fitness
Incorporate diversity into the fitness, where you count how many individuals have the same fitness value (thus are likely to have the same phenotype) and you penalize their fitness by a proportionate value: the more individuals with the same fitness value, the more penalty for those individuals. This way specimens with unique phenotypes will be encouraged, therefore there will be much less stagnation of the population.
Those are just some of the things that can greatly affect the performance of a CGP, and when I say greatly I mean that it's in the same order or greater than Ferriera's performance. So if Ferriera didn't tinker with those ideas too much, then she could have seen much slower performance of the CGPs... especially if she didn't do anything to combat stagnation. So I would be careful when reading performance statistics on GEP, because sometimes people fail to account for all of the "optimizations" available out there.
There seems to be some confusion in these answers that must be clarified. Cartesian GP is different from classic GP (aka tree-based GP), and GEP. Even though they share many concepts and take inspiration from the same biological mechanisms, the representation of the individuals (the solutions) varies.
In CGPthe representation (mapping between genotype and phenotype) is indirect, in other words, not all of the genes in a CGP genome will be expressed in the phenome (a concept also found in GEP and many others). The genotypes can be coded in a grid or array of nodes, and the resulting program graph is the expression of active nodes only.
In GEP the representation is also indirect, and similarly not all genes will be expressed in the phenotype. The representation in this case is much different from treeGP or CGP, but the genotypes are also expressed into a program tree. In my opinion GEP is a more elegant representation, easier to implement, but also suffers from some defects like: you have to find the appropriate tail and head size which is problem specific, the mnltigenic version is a bit of a forced glue between expression trees, and finally it has too much bloat.
Independently of which representation may be better than the other in some specific problem domain, they are general purpose, can be applied to any domain as long as you can encode it.
In general, GEP is simpler from GP. Let's say you allow the following nodes in your program: constants, variables, +, -, *, /, if, ...
For each of such nodes with GP you must create the following operations:
- randomize
- mutate
- crossover
- and probably other genetic operators as well
In GEP for each of such nodes only one operation is needed to be implemented: deserialize, which takes array of numbers (like double in C or Java), and returns the node. It resembles object deserialization in languages like Java or Python (the difference is that deserialization in programming languages uses byte arrays, where here we have arrays of numbers). Even this 'deserialize' operation doesn't have to be implemented by the programmer: it can be implemented by a generic algorithm, just like it's done in Java or Python deserialization.
This simplicity from one point of view may make searching of best solution less successful, but from other side: requires less work from programmer and simpler algorithms may execute faster (easier to optimize, more code and data fits in CPU cache, and so on). So I would say that GEP is slightly better, but of course the definite answer depends on problem, and for many problems the opposite may be true.
Today I read this blog entry by Roger Alsing about how to paint a replica of the Mona Lisa using only 50 semi transparent polygons.
I'm fascinated with the results for that particular case, so I was wondering (and this is my question): how does genetic programming work and what other problems could be solved by genetic programming?
There is some debate as to whether Roger's Mona Lisa program is Genetic Programming at all. It seems to be closer to a (1 + 1) Evolution Strategy. Both techniques are examples of the broader field of Evolutionary Computation, which also includes Genetic Algorithms.
Genetic Programming (GP) is the process of evolving computer programs (usually in the form of trees - often Lisp programs). If you are asking specifically about GP, John Koza is widely regarded as the leading expert. His website includes lots of links to more information. GP is typically very computationally intensive (for non-trivial problems it often involves a large grid of machines).
If you are asking more generally, evolutionary algorithms (EAs) are typically used to provide good approximate solutions to problems that cannot be solved easily using other techniques (such as NP-hard problems). Many optimisation problems fall into this category. It may be too computationally-intensive to find an exact solution but sometimes a near-optimal solution is sufficient. In these situations evolutionary techniques can be effective. Due to their random nature, evolutionary algorithms are never guaranteed to find an optimal solution for any problem, but they will often find a good solution if one exists.
Evolutionary algorithms can also be used to tackle problems that humans don't really know how to solve. An EA, free of any human preconceptions or biases, can generate surprising solutions that are comparable to, or better than, the best human-generated efforts. It is merely necessary that we can recognise a good solution if it were presented to us, even if we don't know how to create a good solution. In other words, we need to be able to formulate an effective fitness function.
Some Examples
Travelling Salesman
Sudoku
EDIT: The freely-available book, A Field Guide to Genetic Programming, contains examples of where GP has produced human-competitive results.
Interestingly enough, the company behind the dynamic character animation used in games like Grand Theft Auto IV and the latest Star Wars game (The Force Unleashed) used genetic programming to develop movement algorithms. The company's website is here and the videos are very impressive:
http://www.naturalmotion.com/euphoria.htm
I believe they simulated the nervous system of the character, then randomised the connections to some extent. They then combined the 'genes' of the models that walked furthest to create more and more able 'children' in successive generations. Really fascinating simulation work.
I've also seen genetic algorithms used in path finding automata, with food-seeking ants being the classic example.
Genetic algorithms can be used to solve most any optimization problem. However, in a lot of cases, there are better, more direct methods to solve them. It is in the class of meta-programming algorithms, which means that it is able to adapt to pretty much anything you can throw at it, given that you can generate a method of encoding a potential solution, combining/mutating solutions, and deciding which solutions are better than others. GA has an advantage over other meta-programming algorithms in that it can handle local maxima better than a pure hill-climbing algorithm, like simulated annealing.
I used genetic programming in my thesis to simulate evolution of species based on terrain, but that is of course the A-life application of genetic algorithms.
The problems GA are good at are hill-climbing problems. Problem is that normally it's easier to solve most of these problems by hand, unless the factors that define the problem are unknown, in other words you can't achieve that knowledge somehow else, say things related with societies and communities, or in situations where you have a good algorithm but you need to fine tune the parameters, here GA are very useful.
A situation of fine tuning I've done was to fine tune several Othello AI players based on the same algorithms, giving each different play styles, thus making each opponent unique and with its own quirks, then I had them compete to cull out the top 16 AI's that I used in my game. The advantage was they were all very good players of more or less equal skill, so it was interesting for the human opponent because they couldn't guess the AI as easily.
http://en.wikipedia.org/wiki/Genetic_algorithm#Problem_domains
You should ask yourself : "Can I (a priori) define a function to determine how good a particular solution is relative to other solutions?"
In the mona lisa example, you can easily determine if the new painting looks more like the source image than the previous painting, so Genetic Programming can be "easily" applied.
I have some projects using Genetic Algorithms. GA are ideal for optimization problems, when you cannot develop a fully sequential, exact algorithm do solve a problem. For example: what's the best combination of a car characteristcs to make it faster and at the same time more economic?
At the moment I'm developing a simple GA to elaborate playlists. My GA has to find the better combinations of albums/songs that are similar (this similarity will be "calculated" with the help of last.fm) and suggests playlists for me.
There's an emerging field in robotics called Evolutionary Robotics (w:Evolutionary Robotics), which uses genetic algorithms (GA) heavily.
See w:Genetic Algorithm:
Simple generational genetic algorithm pseudocode
Choose initial population
Evaluate the fitness of each individual in the population
Repeat until termination: (time limit or sufficient fitness achieved)
Select best-ranking individuals to reproduce
Breed new generation through crossover and/or mutation (genetic
operations) and give birth to
offspring
Evaluate the individual fitnesses of the offspring
Replace worst ranked part of population with offspring
The key is the reproduction part, which could happen sexually or asexually, using genetic operators Crossover and Mutation.