Introduction to valuation of black and white games templates

There are three very important concepts about the valuation of black and white games: mobility, potential mobility, and stability. Another concept is parity, which plays a role in the final stage, it was useless at the beginning. If you do not know this, you can search for it online. This document describes the principles of template valuation.
The calculation of these parameters is complex. In order to accelerate the valuation, a template estimation method is proposed. The idea of template valuation is to transform the global mobility, potential mobility and stability sub-into local mobility, potential mobility and stability sub-, and then combine these local parameters to represent global parameters. Each part contains a small number of pieces, which can be calculated in advance, so that the final valuation can be replaced by a look-up table, so as to speed up.
A template is an area on a board, such as a line on the board and a 3-square block on the top of a corner. Note: The first and eighth rows on the board, the first and eighth columns are equivalent and belong to the same template because they can be obtained through rotation. Likewise, the blocks in the four corners are also the same template. Each player in the template has three statuses: black, white, and empty. Therefore, a template Containing N pawns has a total of three ^ N statuses, also known as template configuration. Based on the local mobility, potential mobility, and stability that each template configuration can represent, You can rate each template configuration. when evaluating a situation, you can split the game board into templates, find the score of each part based on the configuration of each template, and add up to get the score of the entire situation.
I don't know whether there is any theoretical guidance for the Template Selection, but most powerful programs have made similar choices. Taking zebra as an example, he chose 11 templates, the Rows Are -- 2nd, 3, and 4, with a length of 4, 5, 6, 7, and 8. The first row is added with two X cells (X is the location of B2 ), square with a 3-side length on the corner and a 2x5 rectangle on the corner. There are two types of slashes with a length of 8 through rotation, while the two-X5 rectangular blocks have eight types of images in addition to rotation, and the other are four types of instances. Therefore, we divide a situation into 46 templates, and finally sum up the scores of these 46 templates to get the total score of the situation.
The template coefficient is the score configured for each template. The total number of template configurations is huge. The 11 templates mentioned above have a total of over 10 million configurations, and there are great differences between different stages, therefore, it is necessary to divide the entire reconciliation process into multiple stages. It is impossible to score them manually. An algorithm must be used to automatically score the score. Fortunately, it is not difficult for black and white games. You can download a lot of board games from the Internet. Each Board contains records of every step from the start to the end, so that you can calculate the final result of this board. Analyze a large number of board games (these games are called training board games), obtain 46 template configurations for each board, and then approach the final result, in this way, the score of the template configuration encountered in the Training Board is calculated. Of course, there must be many template configurations that are never encountered in the training board. These template configurations will give a neutral 0 score. For the specific calculation method, please refer to http://www.cs.ualberta.ca /~ Mburo/PS/improve.pdf, which is not written due to the inconvenience of inputting mathematical symbols. To calculate the template coefficient, you must note that the template configuration is symmetric. In the above 11 templates, except for the 2x5 template, Other templates have symmetric configuration problems. For example, the "bbww" configuration is equivalent to the "wwbb" configuration, and the two symmetric configurations may have different scores in the result calculated by the Training Board, my solution is to take the weighted average value for the two symmetric configurations to make them equal, but I am not sure whether this method is reasonable.
Template valuation effect. The template coefficient is calculated based on the statistics of the existing game board. Its effect is closely related to the selected training game board, the Training Board should cover all possible board games as much as possible, and usually think that you should choose a board game with a master. Note that you do not require both sides to be a master, which leads to poor coverage. In this way, the calculation result of the template coefficient will reflect the actual situation more realistically. The closer the template valuation is to the final conclusion, the more accurate it is, and the greater the error at the start, which is a weakness of the template valuation. Also, template configurations that are never encountered in the Training Board, because the coefficient is 0, if encountered in the actual game, it will also bring errors.
The training board I selected may not work well because I chose a board where both sides are experts. In practice, in the same search depth, it is difficult to directly use the power, potential mobility, and stability sub-valuation procedures. For template configurations that have never been encountered in the Training Board, I want to re-train the trained template. The specific method is under consideration and will be continued ......