I have a full-time hero system

Chapter 181 How Go Software Works

Yuan Xiao is sure to convince Jiang Zhanhong.

What archaeologists dig up, if no one knows how to use it, is waste.

For some reason, Yuan Xiao thought that Jiang Zhanhong would not give up his discovery so easily.

Sure enough, after a few minutes of silence, Jiang Zhanhong on the other end of the phone agreed.

He set off at once for Rirenwell.

After hanging up the phone, Ke Jie came over: "Teacher Yuan, why don't you tell us the general idea of ​​this Go software."

God knows how he suffered these days.

A software program that can guide chess players to play chess, everyone wants.

It's like reading a book and seeing a wonderful place, but the author doesn't update it.

Ke Jie felt itchy for Yuan Xiao.

Students are also interested, after all, everyone likes new things.

Yuan Xiao simply sat down and let Zhuo Liang fiddle with the computer.

"Any game of chess, under the rules, mostly tests calculation power." Yuan Xiao began to explain.

If you want to play poker well, even if it is fighting the landlord, you need to calculate it.

Chess, checkers, backgammon and so on.

Including Go, in addition to some corresponding skills, it is actually a test of computing power.

Ke Jie admitted: "This is also the reason why the current computer industry thinks that Go software is difficult to develop."

The reason why it is difficult is that there are too many points to choose from in Go.

A 19-way chessboard has 19 horizontal and vertical lines, a total of 361 points.

In other words, there are theoretically 361 possibilities for where to go in the first step.

The second step is 360 possibilities.

And so on.

"I checked the information just now." Su Chenyi turned the computer screen over, "For example, chess software is actually using the method of exhaustion.

Simply put, count all possibilities. "

The opening move, the result has been predicted, of course, it will not lose.

This is where the computer is better than the human brain.

"But Go can't," Ke Jie added, "If you want to exhaust all possibilities, that's an astronomical figure."

"Yes," Yuan Xiao was also very happy, "So, we can't use this type of algorithm."

This is where the Monte Carlo algorithm is needed.

This algorithm is actually a calculation of a sampling sample.

For example, find the largest apple from a box of apples.

What normal people do is to take out an apple, compare it with another one, and keep the bigger one.

This has been eliminated until the last remaining apple is the largest.

Then, the result of this processing depends on how many apples are in the box.

The more words, the more accurate the results will be.

From this aspect, it is very close to the thinking of playing Go.

Ke Jie understood.

When a master plays Go, he also simulates and calculates the next few steps based on the current situation, and chooses the optimal plan.

"That is to say," Ke Jie squeezed his chin, "it is not necessary to calculate all the possibilities, but to select some important ones for analysis."

"Yes, this is the core point of the Go software program." Yuan Xiao raised his finger.

The computer analyzes every move in this way.

Of course, the point of analysis is definitely not the number of hundreds and thousands.

Su Chenyi also understood: "The key to the problem is which points need to be analyzed and which points can be eliminated directly."

Yuan Xiao was very satisfied, and Xiao Chenyi was really awesome.

Fang Xuyao ​​is not to be outdone, after all he is the person with the highest IQ in the country: "It's a bit like pruning."

"That's right." Talking to smart people saves trouble, Yuan Xiao smiled, "It's just pruning."

A bonsai with a branched trunk and many branches.

What Yuan Xiao has to do is to cut out the unnecessary ones and keep the useful ones.

Through this method, the scope and amount of calculation are compressed, so that the program can proceed smoothly.

Ke Jie nodded: "The replacement of the concept is to get the optimal solution of the current situation, not the final solution of the whole game."

The two are worlds apart.

However, as long as from the beginning, every move played by the computer is the current optimal solution, and when accumulated, it becomes the optimal solution of the whole chess game.

Proud index +8888
Look, Ke Jie was shocked by just doing it a few times.

It's a pity that most of the students don't like Yuan Xiao.

This is where it is difficult to collect the Aoshi Index.

It seems that only when the finished product comes out, can they harvest a wave.

There are pitfalls in the task.

Ke Jie was very interested: "Then the question is how to screen out such points."

"So, we need to add to the algorithm." Yuan Xiao explained.

A point on the Go board can be represented by a numerical value.

For example, if there is a white piece, it is 1, if there is a black piece, it is -1, and if there is no piece, it is 0.

This is the most basic.

In this way, mathematical values ​​can be established for the 361 points on the entire chessboard.

Then, the program can perform repeated calculations on this group of values.

Yuan Xiao drew a conceptual diagram: "The first thing we need is a strategy neural network."

The role of the strategy neural network is to judge where the next move is most appropriate in the current situation.

Yuan Xiao believes that the selection of points here can refer to the game records of human chess players.

Imitate the thinking mode of human chess players, and quickly eliminate some low-probability points.

Ke Jie sighed: "It seems that at that time, I will have to provide all the game records of all human beings so far."

In theory, the greater the number of chess records, the higher the maturity of the strategy neural network.

"At this point, we let the computer learn independently." Yuan Xiao pointed out the key.

The results of repeated deduction and learning will make the strategy neural network more powerful and the selection of points more accurate.

"To achieve this step, another combination of neural networks is needed." Yuan Xiao wrote down a formula.

value neural network.

Its function is to be able to calculate the winning rate of each selected point.

Such valuation judgments are carried out at each step until the end of the game to obtain the most accurate winning percentage value.

And again, this thing can also be learned by yourself.

It is actually very simple to sum up.

It is to simulate the thinking of human playing chess.

First of all, judge where you can play, and what is the winning rate of each place.

Then perform centralized calculation on these selected points to deduce the final result.

Do this every step of the way.

Then every move you get is naturally the best way to play.

If it is said that Ke Jie thought Yuan Xiao was talking about the Arabian Nights at the beginning.

Now, he feels that this thing is not completely impossible.

He swallowed: "In this way, the computer has already locked the result since the beginning."

A must-win result.

Proud index +8888
"This is just a general concept. In reality, we still need to take it step by step," Yuan Xiao put down his pen, "Dean Ke, when the time comes to play against the Go software, you will be needed."

Ke Jie expressed his duty.

Su Chenyi, Zhuo Zhuo, and Fang Xuyao ​​all understood.

Of course Xia Wanqing, Li Mulan and her like don't know what to say.

Especially Xia Wanqing, she expressed her contempt.

What's the use of spending your brain to develop this thing.

The teacher is also really, wouldn't it be nice to spend time with me?

The world of computers really contains the attributes of nerds.

　Thanks to @雷雨下的小草 for 2 monthly tickets, New Leaf 0000, and Dog Uncle Guhan for their support!
　　Set up a bowl for votes!

　　
　
(End of this chapter)