Test results after Mobility is implemented

[calcitem]

Modified on the basis of v1.0.23, in the placing phase and the moving phase , directly add the Mobility value, that is, the difference between the pieces * 5 + Mobility, and note that the search depth expansion is not performed when the need remove is non-zero. Test Results:

With DevMode turned on, play 2300 games, the modified first move: [23.0%] : 20.0% : 57.1%, the second move, 8.8% : [40.0%] : 51.2%, obviously dominant, but it may also be due to the opening The key is that the choices are different.
Then, based on the version that calculates the difference in mobility in real time, we played against 2000-2500 games. The results are as follows:
White: [22.5%] : 19.7% : 57.8%
Black: 9.0% : [42.9%] : 48.2%
Indeed a significant improvement.

With DevMode turned on, self-play, 7605 games were played after the improvement, 3976 games were played before the improvement, and the efficiency search was significantly increased after the improvement. The reason is unknown.
After the improvement, the self-play ratio is 11.5% : 27.1% : 61.5%, and before the improvement, the self-play ratio is 20.0% : 22.6 % : 57.4%.

For example, the search depth is 20, the thinking time is 1s, and the results of the 700-900 game after and before the improvement are as follows:
Improved white: [10.3%] : 37.9% : 51.8%;
After the improvement, the black holding: 9.9% : [26.4%] : 63.7%, that is, the black holding became weaker after the improvement.
Then, based on the version that calculates the difference in mobility in real time, we played against 2000-2500 games. The results are as follows:
Improved white: [8.3%] : 33.1% : 58.6%;
After the improvement, the black holding: 10.7% : [24.2%] : 65.1%, that is, the black holding after the improvement has indeed become weaker.

Playing 1700-5000 games against Perfect AI, no matter before or after improvement, the draw rate for white is 36.6%, and the rate for black is 53%-54%.