I am currently estimating a Markov-switching model with many parameters using direct optimization of the log likelihood function (through the forward-backward algorithm). I do the numerical optimization using matlab's genetic algorithm, since other approaches such as the (mostly gradient or simplex-based) algorithms in fmincon and fminsearchbnd were not very useful, given that likelihood function is not only of very high dimension but also shows many local maxima and is highly nonlinear.
The genetic algorithm seems to work very well. However, I am planning to further increase the dimension of the problem. I have read about an EM algorithm to estimate Markov-switching models. From what I understand this algorithm releases a sequence of increasing log-likelhood values. It thus seems suitable to estimate models with very many parameters.
My question is if the EM algorithm is suitable for my application involving many parameters (perhaps better suitable as the genetic algorithm). Speed is not the main limitation (the genetic algorithm is altready extremely slow) but I would need to have some certainty to end up close to the global optimum and not run into one of the many local optima. Do you have any experience or suggestions regarding this?
The EM algorithm finds local optima, and does not guarantee that they are global optima. In fact, if you start it off with a HMM where one of the transition probabilities is zero, that probability will typically never change from zero, because those transitions will appear only with expectation zero in the expectation step, so those starting points have no hope of finding a global optimum which does not have that transition probability zero.
The standard workaround for this is to start it off from a variety of different random parameter settings, pick the highest local optima found, and hope for the best. You might be slightly reassured if a significant proportion of the runs converged to the same (or to equivalent) best local optimum found, on the not very reliable theory that anything better would be found from at least the same fraction of random starts, and so would have showed up by now.
I haven't worked it out in detail, but the EM algorithm solves such a general set of problems that I expect that if it guaranteed to find the global optimum then it would be capable of finding the solution to NP-complete problems with unprecedented efficiency.
Related
I am learning about the k-means clustering algorithm, and I have read that the algorithm is "Trying to minimise a loss function in which the goal of clustering is not met".
I understand the basic concept of the algorithm, which initialises arbitrary centroids/means in the first iteration and then assigns data points to these clusters. The centroids are then updated after the points are all assigned, and points are re-assigned again. The algorithm continues to iterate until the clusters do not change anymore. The algorithm tries to minimise the within-cluster sum of squares (WCSS) value which is a measure of the variance within the clusters.
However, I am having trouble understanding what is meant by a loss function in the context of this algorithm. Any insights are appreciated.
In the broader context of machine learning, a loss function (sometimes called a cost function) is a function that indicates how "bad" a particular answer is. Minimizing the loss function then corresponds to finding the "best" answer to the problem.
(It may be weird to think about "losing the least amount," in which case thinking of it as a cost function might be easier. The goal is to then find the lowest-cost solution to the problem.)
Currently, I'm studying genetic algorithms (personal, not required) and I've come across some topics I'm unfamiliar or just basically familiar with and they are:
Search Space
The "extreme" of a Function
I understand that one's search space is a collection of all possible solutions but I also wish to know how one would decide the range of their search space. Furthermore I would like to know what an extreme is in relation to functions and how it is calculated.
I know I should probably understand what these are but so far I've only taken Algebra 2 and Geometry but I have ventured into physics, matrix/vector math, and data structures on my own so please excuse me if I seem naive.
Generally, all algorithms which are looking for a specific item in a collection of items are called search algorithms. When the collection of items is defined by a mathematical function (opposed to existing in a database), it is called a search space.
One of the most famous problems of this kind is the travelling salesman problem, where an algorithm is sought which will, given a list of cities and their distances, find the shortest route for visiting each city only once. For this problem, the exact solution can be found only by examining all possible routes (the entire search space), and finding the shortest one (the route which has the minimum distance, which is the extreme value in the search space). The best time complexity of such an algorithm (called an exhaustive search) is exponential (although it is still possible that there may be a better solution), meaning that the worst-case running time increases exponentially as the number of cities increases.
This is where genetic algorithms come into play. Similar to other heuristic algorithms, genetic algorithms try to get close to the optimal solution by improving a candidate solution iteratively, with no guarantee that an optimal solution will actually be found.
This iterative approach has the problem that the algorithm can easily get "stuck" in a local extreme (while trying to improve a solution), not knowing that there is a potentially better solution somewhere further away:
The figure shows that, in order to get to the actual, optimal solution (the global minimum), an algorithm currently examining the solution around the local minimum needs to "jump over" a large maximum in the search space. A genetic algorithm will rapidly locate such local optimums, but it will usually fail to "sacrifice" this short-term gain to get a potentially better solution.
So, a summary would be:
exhaustive search
examines the entire search space (long time)
finds global extremes
heuristic (e.g. genetic algorithms)
examines a part of the search space (short time)
finds local extremes
Genetic algorithms are not good in tuning to a local optimum. If you want to find a global optimum at least you should be able to approach or find a strategy to approach the local optimum. Recently some improvements have been developed to better find the local optima.
"GENETIC ALGORITHM FOR INFORMATIVE BASIS FUNCTION SELECTION
FROM THE WAVELET PACKET DECOMPOSITION WITH APPLICATION TO
CORROSION IDENTIFICATION USING ACOUSTIC EMISSION"
http://gbiomed.kuleuven.be/english/research/50000666/50000669/50488669/neuro_research/neuro_research_mvanhulle/comp_pdf/Chemometrics.pdf
In general, "search space" means, what type of answers are you looking for. For example, if you are writing a genetic algorithm which builds bridges, tests them out, and then builds more, the answers you are looking for are bridge models (in some form). As another example, if you're trying to find a function which agrees with a set of sample inputs on some number of points, you might try to find a polynomial which has this property. In this instance your search space might be polynomials. You might make this simpler by putting a bound on the number of terms, maximum degree of the polynomial, etc... So you could specify that you wanted to search for polynomials with integer exponents in the range [-4, 4]. In genetic algorithms, the search space is the set of possible solutions you could generate. In genetic algorithms you need to carefully limit your search space so you avoid answers which are completely dumb. At my former university, a physics student wrote a program which was a GA to calculate the best configuration of atoms in a molecule to have low energy properties: they found a great solution having almost no energy. Unfortunately, their solution put all the atoms at the exact center of the molecule, which is physically impossible :-). GAs really hone in on good solutions to your fitness functions, so it's important to choose your search space so that it doesn't produce solutions with good fitness but are in reality "impossible answers."
As for the "extreme" of a function. This is simply the point at which the function takes its maximum value. With respect to genetic algorithms, you want the best solution to the problem you're trying to solve. If you're building a bridge, you're looking for the best bridge. In this scenario, you have a fitness function that can tell you "this bridge can take 80 pounds of weight" and "that bridge can take 120 pounds of weight" then you look around for solutions which have higher fitness values than others. Some functions have simple extremes: you can find the extreme of a polynomial using simple high school calculus. Other functions don't have a simple way to calculate their extremes. Notably, highly nonlinear functions have extremes which might be difficult to find. Genetic algorithms excel at finding these solutions using a clever search technique which looks around for high points and then finds others. It's worth noting that there are other algorithms that do this as well, hill climbers in particular. The things that make GAs different is that if you find a local maximum, other types of algorithms can get "stuck," blinded by a locally good solution, so that they never see a possibly much better solution farther away in the search space. There are other ways to adapt hill climbers to this as well, simulated annealing, for one.
The range space usually requires some intuitive understanding of the problem you're trying to solve-- some expertise in the domain of the problem. There's really no guaranteed method to pick the range.
The extremes are just the minimum and maximum values of the function.
So for instance, if you're coding up a GA just for practice, to find the minimum of, say, f(x) = x^2, you know pretty well that your range should be +/- something because you already know that you're going to find the answer at x=0. But then of course, you wouldn't use a GA for that because you already have the answer, and even if you didn't, you could use calculus to find it.
One of the tricks in genetic algorithms is to take some real-world problem (often an engineering or scientific problem) and translate it, so to speak, into some mathematical function that can be minimized or maximized. But if you're doing that, you probably already have some basic notion where the solutions might lie, so it's not as hopeless as it sounds.
The term "search space" does not restrict to genetic algorithms. I actually just means the set of solutions to your optimization problem. An "extremum" is one solution that minimizes or maximizes the target function with respect to the search space.
Search space simply put is the space of all possible solutions. If you're looking for a shortest tour, the search space consists of all possible tours that can be formed. However, beware that it's not the space of all feasible solutions! It only depends on your encoding. If your encoding is e.g. a permutation, than the search space is that of the permutation which is n! (factorial) in size. If you're looking to minimize a certain function the search space with real valued input the search space is bounded by the hypercube of the real valued inputs. It's basically infinite, but of course limited by the precision of the computer.
If you're interested in genetic algorithms, maybe you're interested in experimenting with our software. We're using it to teach heuristic optimization in classes. It's GUI driven and windows based so you can start right away. We have included a number of problems such as real-valued test functions, traveling salesman, vehicle routing, etc. This allows you to e.g. look at how the best solution of a certain TSP is improving over the generations. It also exposes the problem of parameterization of metaheuristics and lets you find better parameters that will solve the problems more effectively. You can get it at http://dev.heuristiclab.com.
Recently I've been looking at some greedy algorithm problems. I am confused about locally optimal. As you know, greedy algorithms are composed of locally optimal choices. But combining of locally optimal decisions doesn't necessarily mean globally optimal, right?
Take making change as an example: using the least number of coins to make 15¢, if we have
10¢, 5¢, and 1¢ coins then you can achieve this with one 10¢ and one 5¢. But if we add in a 12¢ coin the greedy algorithm fails as (1×12¢ + 3×1¢) uses more coins than (1×10¢ + 1×5¢).
Consider some classic greedy algorithms, e.g. Huffman, Dijkstra. In my opinion, these algorithms are successful as they have no degenerate cases which means a combination of locally optimal steps always equals global optimal. Do I understand right?
If my understanding is correct, is there a general method for checking if a greedy algorithm is optimal?
I found some discussion of greedy algorithms elsewhere on the site.
However, the problem doesn't go into too much detail.
Generally speaking, a locally optimal solution is always a global optimum whenever the problem is convex. This includes linear programming; quadratic programming with a positive definite objective; and non-linear programming with a convex objective function. (However, NLP problems tend to have a non-convex objective function.)
Heuristic search will give you a global optimum with locally optimum decisions if the heuristic function has certain properties. Consult an AI book for details on this.
In general, though, if the problem is not convex, I don't know of any methods for proving global optimality of a locally optimal solution.
There are some theorems that express problems for which greedy algorithms are optimal in terms of matroids (also:greedoids.) See this Wikipedia section for details: http://en.wikipedia.org/wiki/Matroid#Greedy_algorithms
A greedy algorithm almost never succeeds in finding the optimal solution. In the cases that it does, this is highly dependent on the problem itself. As Ted Hopp explained, with convex curves, the global optimal can be found, assuming you are to find the maximum of the objective function of course (conversely, concave curves also work if you are to minimise). Otherwise, you will almost certainly get stuck in the local optima. This assumes that you already know the objective function.
Another factor which I can think of is the neighbourhood function. Certain neighbourhoods, if large enough, will encompass both the global and local maximas, so that you can avoid the local maxima. However, you can't make the neighbourhood too large or search will be slow.
In other words, whether you find a global optimal or not with greedy algorithms is problem specific, although for most cases, you will not find the globally optimal.
You need to design a witness example where your premise that the algorithm is a global one fails. Design it according to the algorithm and the problem.
Your example of the coin change was not a valid one. Coins are designed purposely to have all the combinations possible, but not to add confusion. Your addition of 12c is not warranted and is extra.
With your addition, the problem is not coin change but a different one (even though the subject are coins, you can change the example to what you want). For this, you yourself gave a witness example to show the greedy algorithm for this problem will get stuck in a local maximum.
Has anyone tried to apply a smoother to the evaluation metric before applying the L-method to determine the number of k-means clusters in a dataset? If so, did it improve the results? Or allow a lower number of k-means trials and hence much greater increase in speed? Which smoothing algorithm/method did you use?
The "L-Method" is detailed in:
Determining the Number of Clusters/Segments in Hierarchical Clustering/Segmentation Algorithms, Salvador & Chan
This calculates the evaluation metric for a range of different trial cluster counts. Then, to find the knee (which occurs for an optimum number of clusters), two lines are fitted using linear regression. A simple iterative process is applied to improve the knee fit - this uses the existing evaluation metric calculations and does not require any re-runs of the k-means.
For the evaluation metric, I am using a reciprocal of a simplified version of the Dunns Index. Simplified for speed (basically my diameter and inter-cluster calculations are simplified). The reciprocal is so that the index works in the correct direction (ie. lower is generally better).
K-means is a stochastic algorithm, so typically it is run multiple times and the best fit chosen. This works pretty well, but when you are doing this for 1..N clusters the time quickly adds up. So it is in my interest to keep the number of runs in check. Overall processing time may determine whether my implementation is practical or not - I may ditch this functionality if I cannot speed it up.
I had asked a similar question in the past here on SO. My question was about coming up with a consistent way of finding the knee to the L-shape you described. The curves in question represented the trade-off between complexity and a fit measure of the model.
The best solution was to find the point with the maximum distance d according to the figure shown:
Note: I haven't read the paper you linked to yet..
What is the difference between a heuristic and an algorithm?
An algorithm is the description of an automated solution to a problem. What the algorithm does is precisely defined. The solution could or could not be the best possible one but you know from the start what kind of result you will get. You implement the algorithm using some programming language to get (a part of) a program.
Now, some problems are hard and you may not be able to get an acceptable solution in an acceptable time. In such cases you often can get a not too bad solution much faster, by applying some arbitrary choices (educated guesses): that's a heuristic.
A heuristic is still a kind of an algorithm, but one that will not explore all possible states of the problem, or will begin by exploring the most likely ones.
Typical examples are from games. When writing a chess game program you could imagine trying every possible move at some depth level and applying some evaluation function to the board. A heuristic would exclude full branches that begin with obviously bad moves.
In some cases you're not searching for the best solution, but for any solution fitting some constraint. A good heuristic would help to find a solution in a short time, but may also fail to find any if the only solutions are in the states it chose not to try.
An algorithm is typically deterministic and proven to yield an optimal result
A heuristic has no proof of correctness, often involves random elements, and may not yield optimal results.
Many problems for which no efficient algorithm to find an optimal solution is known have heuristic approaches that yield near-optimal results very quickly.
There are some overlaps: "genetic algorithms" is an accepted term, but strictly speaking, those are heuristics, not algorithms.
Heuristic, in a nutshell is an "Educated guess". Wikipedia explains it nicely. At the end, a "general acceptance" method is taken as an optimal solution to the specified problem.
Heuristic is an adjective for
experience-based techniques that help
in problem solving, learning and
discovery. A heuristic method is used
to rapidly come to a solution that is
hoped to be close to the best possible
answer, or 'optimal solution'.
Heuristics are "rules of thumb",
educated guesses, intuitive judgments
or simply common sense. A heuristic is
a general way of solving a problem.
Heuristics as a noun is another name
for heuristic methods.
In more precise terms, heuristics
stand for strategies using readily
accessible, though loosely applicable,
information to control problem solving
in human beings and machines.
While an algorithm is a method containing finite set of instructions used to solving a problem. The method has been proven mathematically or scientifically to work for the problem. There are formal methods and proofs.
Heuristic algorithm is an algorithm that is able to produce an
acceptable solution to a problem in
many practical scenarios, in the
fashion of a general heuristic, but
for which there is no formal proof of
its correctness.
An algorithm is a self-contained step-by-step set of operations to be performed 4, typically interpreted as a finite sequence of (computer or human) instructions to determine a solution to a problem such as: is there a path from A to B, or what is the smallest path between A and B. In the latter case, you could also be satisfied with a 'reasonably close' alternative solution.
There are certain categories of algorithms, of which the heuristic algorithm is one. Depending on the (proven) properties of the algorithm in this case, it falls into one of these three categories (note 1):
Exact: the solution is proven to be an optimal (or exact solution) to the input problem
Approximation: the deviation of the solution value is proven to be never further away from the optimal value than some pre-defined bound (for example, never more than 50% larger than the optimal value)
Heuristic: the algorithm has not been proven to be optimal, nor within a pre-defined bound of the optimal solution
Notice that an approximation algorithm is also a heuristic, but with the stronger property that there is a proven bound to the solution (value) it outputs.
For some problems, noone has ever found an 'efficient' algorithm to compute the optimal solutions (note 2). One of those problems is the well-known Traveling Salesman Problem. Christophides' algorithm for the Traveling Salesman Problem, for example, used to be called a heuristic, as it was not proven that it was within 50% of the optimal solution. Since it has been proven, however, Christophides' algorithm is more accurately referred to as an approximation algorithm.
Due to restrictions on what computers can do, it is not always possible to efficiently find the best solution possible. If there is enough structure in a problem, there may be an efficient way to traverse the solution space, even though the solution space is huge (i.e. in the shortest path problem).
Heuristics are typically applied to improve the running time of algorithms, by adding 'expert information' or 'educated guesses' to guide the search direction. In practice, a heuristic may also be a sub-routine for an optimal algorithm, to determine where to look first.
(note 1): Additionally, algorithms are characterised by whether they include random or non-deterministic elements. An algorithm that always executes the same way and produces the same answer, is called deterministic.
(note 2): This is called the P vs NP problem, and problems that are classified as NP-complete and NP-hard are unlikely to have an 'efficient' algorithm. Note; as #Kriss mentioned in the comments, there are even 'worse' types of problems, which may need exponential time or space to compute.
There are several answers that answer part of the question. I deemed them less complete and not accurate enough, and decided not to edit the accepted answer made by #Kriss
Actually I don't think that there is a lot in common between them. Some algorithm use heuristics in their logic (often to make fewer calculations or get faster results). Usually heuristics are used in the so called greedy algorithms.
Heuristics is some "knowledge" that we assume is good to use in order to get the best choice in our algorithm (when a choice should be taken). For example ... a heuristics in chess could be (always take the opponents' queen if you can, since you know this is the stronger figure). Heuristics do not guarantee you that will lead you to the correct answer, but (if the assumptions is correct) often get answer which are close to the best in much shorter time.
An Algorithm is a clearly defined set of instructions to solve a problem, Heuristics involve utilising an approach of learning and discovery to reach a solution.
So, if you know how to solve a problem then use an algorithm. If you need to develop a solution then it's heuristics.
Heuristics are algorithms, so in that sense there is none, however, heuristics take a 'guess' approach to problem solving, yielding a 'good enough' answer, rather than finding a 'best possible' solution.
A good example is where you have a very hard (read NP-complete) problem you want a solution for but don't have the time to arrive to it, so have to use a good enough solution based on a heuristic algorithm, such as finding a solution to a travelling salesman problem using a genetic algorithm.
Algorithm is a sequence of some operations that given an input computes something (a function) and outputs a result.
Algorithm may yield an exact or approximate values.
It also may compute a random value that is with high probability close to the exact value.
A heuristic algorithm uses some insight on input values and computes not exact value (but may be close to optimal).
In some special cases, heuristic can find exact solution.
A heuristic is usually an optimization or a strategy that usually provides a good enough answer, but not always and rarely the best answer. For example, if you were to solve the traveling salesman problem with brute force, discarding a partial solution once its cost exceeds that of the current best solution is a heuristic: sometimes it helps, other times it doesn't, and it definitely doesn't improve the theoretical (big-oh notation) run time of the algorithm
I think Heuristic is more of a constraint used in Learning Based Model in Artificial Intelligent since the future solution states are difficult to predict.
But then my doubt after reading above answers is
"How would Heuristic can be successfully applied using Stochastic Optimization Techniques? or can they function as full fledged algorithms when used with Stochastic Optimization?"
http://en.wikipedia.org/wiki/Stochastic_optimization
One of the best explanations I have read comes from the great book Code Complete, which I now quote:
A heuristic is a technique that helps you look for an answer. Its
results are subject to chance because a heuristic tells you only how
to look, not what to find. It doesn’t tell you how to get directly
from point A to point B; it might not even know where point A and
point B are. In effect, a heuristic is an algorithm in a clown suit.
It’s less predict- able, it’s more fun, and it comes without a 30-day,
money-back guarantee.
Here is an algorithm for driving to someone’s house: Take Highway 167
south to Puy-allup. Take the South Hill Mall exit and drive 4.5 miles
up the hill. Turn right at the light by the grocery store, and then
take the first left. Turn into the driveway of the large tan house on
the left, at 714 North Cedar.
Here’s a heuristic for getting to someone’s house: Find the last
letter we mailed you. Drive to the town in the return address. When
you get to town, ask someone where our house is. Everyone knows
us—someone will be glad to help you. If you can’t find anyone, call us
from a public phone, and we’ll come get you.
The difference between an algorithm and a heuristic is subtle, and the
two terms over-lap somewhat. For the purposes of this book, the main
difference between the two is the level of indirection from the
solution. An algorithm gives you the instructions directly. A
heuristic tells you how to discover the instructions for yourself, or
at least where to look for them.
They find a solution suboptimally without any guarantee as to the quality of solution found, it is obvious that it makes sense to the development of heuristics only polynomial. The application of these methods is suitable to solve real world problems or large problems so awkward from the computational point of view that for them there is not even an algorithm capable of finding an approximate solution in polynomial time.