Number Grid Investigation

Digital Grid Survey By digitally multiplying the number in the upper left corner with the number in the lower right corner, you can see the difference between the products. Product differences will be the same as the grid size (width) of 2 x 2 size. In 3 x 3 dimensions, the difference is 4 times the grid (width) size. Also check the operation using 5 x 5 grid, 6 x 6 grid, 7 x 7 grid.

First, check the change in the degree of inversion inversion with respect to the grid number. Each change in the grid lines of a particular group was analyzed and concluded. The graph is rendered by changing the number of grid lines in each group A, B, C, and D for irregular coefficients. It can be seen from these figures that there are different numbers of grid lines in each particular group A, B, C or D, and the maximum results are obtained for that particular group. The maximum value of the irregularity coefficient is the block of the C group where the shear wall is away from the center of gravity instead of the edge. The irregular coefficient decreases when increasing the number of axes after reaching the maximum value of a certain number of grid lines.

The adjacency matrix is ​​a digital grid where each row or column represents a different node in the graph. At the intersection of rows and columns, there is a number representing the relationship. Zero means no edge or relationship. A person means having a relationship. Numbers greater than 1 can be used to display different weights. The traverse algorithm is an algorithm that traverses or accesses nodes in a graph. The main types of traversal algorithms are breadth - first search and depth - first search. One of the uses is to determine how close the node is to the root node. In the video below, let's learn how to implement breadth first search with JavaScript.

The Grid World implementation aims to start the proxy from the upper left corner of the (0,0) grid and reach the lower right corner of the (Ny - 1, Nx - 1) grid angle with minimal steps. Ny + Nx step agent allows only one grid square move up, down, left, and right direction. In order to defeat this game, we used Policy Enhancement Monte Carlo Rewarded Average Sampling Reinforcement Learning Algorithm and used epsilon greedy proxy to navigate the grid world environment. Rewards, Status - In this implementation tabular action values ​​and policies are used (this applies to small grid sizes), and most of the above theory is in Sutton & Barto's "Reinforcement Learning" textbook. The Python code implementation of this algorithm is in Github.