blog posts

What Is Dynamic Programming And Why Is It A Good Idea To Learn It?

When we think of a dynamic programming problem, we need to understand the set of sub-problems and how they relate to each other.

 In this article, we are going to briefly review dynamic programming.

Algorithms and data structures are integral components of data science. However, most data scientists do not properly cover topics related to the analysis and design of algorithms during their studies, while these topics are quite important and not only data scientists but also programmers should have accurate and complete information about By designing algorithms and especially dynamic programming.

In topics related to computer science and statistics, Poil planning and programming is an efficient solution to search and optimization problems using the two features of overlapping subproblems and optimal infrastructure.

Unlike linear programming, there is no standard framework for formulating dynamic programming problems. More specifically, the concept called dynamic planning is to provide a general approach to solving these types of problems.

In each case, special mathematical equations and relations must be developed that match the conditions of the problem.

What is Dynamic Programming?

Dynamic programming often uses an array to store results for reuse, and the subproblems are solved partially as a whole. In this case, the subproblems consisting of only two matrices are first calculated.

Then the sub-problems consisting of three matrices are calculated. These sub-problems are broken down into sub-problems consisting of two matrices that have already been calculated, and the results of which are stored in an array.

As a result, they do not need to be recalculated. Similarly, in the next step, the subproblems are calculated with four matrices, and so on. Dynamic programming, like the division and solution method, solves problems by combining the answers to sub-problems.

The division and solution algorithm divides the problem into independent sub-problems and after solving the subproblems recursively, combines the results and obtains the answer to the main problem.

More specifically, dynamic programming can be used in cases where the sub-problems are not independent; That is when the sub-problems themselves have several identical sub-problems.

In this case, the division and solution method performs more than the required amount of calculations by repeatedly performing the same sub-problems. A dynamic scheduling algorithm stores sub-problems once and stores their answers in a table, thus preventing sub-problems from being repeated when they need to be answered again.

What are the hallmarks of dynamic programming?

An issue must have two key characteristics to be able to use dynamic programming for it. The first is to have an optimal infrastructure and the second is to have overlapping sub-problems. Solving a problem by combining the optimal answers of the overlapping sub-problems is called “division and solving”. This is why fast, integrated sorting is not known as dynamic programming issues.

The key to dynamic programming is the principle of optimality. If the parentheses of the whole phrase are to be optimized, the parentheses of the subproblems must also be optimal. That is, the optimality of the problem requires the optimality of the subproblems.

So you can use the dynamic programming method. The optimal solution is the third step in developing a dynamic programming algorithm for optimization problems. The development steps of such an algorithm are divided into three parts.

Not all optimization problems can be solved with dynamic programming because the principle of optimization must apply to the problem. The principle of optimality applies to a problem if an optimal solution for an instance of the problem always contains the optimal solution for all subsets.

Optimal binary search trees

A binary search tree is a binary tree of elements commonly called a key. Each node contains a key. The keys in the left subtree of each node are less than or equal to the key in that node, and the keys in the right subtree of each node are larger. Or equal to the key of that node

Optimal infrastructure

Optimal infrastructure means that the solution of an optimization problem can be obtained by combining the optimal solutions of its subproblems. Such optimal infrastructures are usually obtained with a recursive approach; For example, in the graph (G = (V, E), the shortest path m from vertex A to vertex B shows the optimal infrastructure: Consider every middle vertex of path m.

If m is really the shortest path, then It can be considered as two sub-paths M1 from A to C and M2 from C to B, so that each of these is the shortest path between the two ends (in the same way as cutting and pasting the book Introduction to Algorithms).

Basically, the answer can be expressed recursively, just like the Belman-Ford and Floyd-Warshall algorithms.

Using the optimal infrastructure method means breaking the problem into smaller sub-problems and finding the optimal answer for each of these sub-problems and obtaining the optimal answer to the whole problem by putting these partial optimal answers together.

For example, when solving the problem of finding the shortest path from one vertex of a graph to another, we can obtain the shortest path to the destination from all adjacent vertices and use it for a general answer.

In general, problem-solving with this method includes three steps of breaking the problem into smaller parts, solving these sub-problems themselves by breaking them backward, and using partial answers to find the general answer.

The sub-problem graph contains the same information for an issue.

Figure 1 shows the sub-problems of the Fibonacci sequences. Sub-Problem Graph A directional graph consists of one vertex for each distinct sub-problem and in the case of a directional edge from the vertex of sub-problem x to the vertex of sub-problem y, which directly requires an optimal solution for sub-problem to determine an optimal solution for sub-problem x. y have.

For example, the graph of the subproblem has an edge from x to y, if a top-down recursive procedure for solving x calls itself directly for solving y.

The subproblem graph can be considered a reduced version of the recursive tree for the top-down recursive method, in which we unite all the vertices of a subproblem and orient the edges from parent to child.

The bottom-up method for dynamic programming considers the outline of the subproblem in such a way that all adjacent subproblems of a subproblem are solved first.

In a dynamic bottom-up programming algorithm, we consider the outline of the subproblems in such a way that the “topologically ordered inverse” or “inverted topologically inverted” is the subset of the problems.

In other words, no subproblem is considered until all the subproblems on which it depends have been resolved.

Similarly, we can see the top-down (along with memorization) approach to dynamic programming in the form of an in-depth graph search of sub-problems.

The graph size of the subproblems can help us determine the execution time of the dynamic programming algorithm. Since we solve each subproblem only once, the execution time is equal to the total number of times we need to solve each subproblem.

Typically, the time to calculate the answer to a sub-problem is proportional to the degree of the corresponding vertex in the graph, and the number of sub-problems is equal to the number of vertices of the graph.

For example, a simple implementation of a function to find the nth Fibonacci number could be as follows.

function fib (n)

if n = 0

return 0

if n = 1

return 1

return fib (n – 1) + fib (n – 2)

One of the most widely used and popular methods of designing algorithms is the Dynamic Programming method.

This method works like the Divide and Conquer method based on dividing the problem into subproblems. But there are significant differences. When a problem is divided into two or more sub-problems, two situations can occur:

The data of the sub-problems have nothing in common and are completely independent of each other. An example of this is the sorting of arrays by the merge method or the fast method, in which the data is divided into two parts and arranged separately.

In this case, the data of one section has nothing to do with the data of the other section and as a result, the result of that section does not affect it. The division and solution method usually works well for such problems.

The data in the sub-issue are related or shared. In this case, the so-called subproblems overlap. A clear example of such problems is the calculation of the nth sentence of the Fibonacci number sequence.

Dynamic programming methods are often used for algorithms that seek to solve problems optimally.

In the method of division and domination, some smaller sub-problems may be equal, in which case the equal sub-problems are solved several times repeatedly, which is one of the disadvantages of the method of division and domination.

Dynamic programming is a bottom-up method, meaning that we go from solving smaller problems to solving bigger problems.

While structurally, the method of division and domination is a top-down method, that is, logically (not practically) the process starts from the initial problem and goes down to solve smaller problems.