4 Experimental Results
4.1 Data set Acquisition
In order to develop the speedometer, we will only consider the Army of each party for
the time being, regardless of the air force situation of the three races, and then consider
expanding the Air Force part after the follow-up algorithm is implemented and verified
[4, 9].
SparCraft requires the following information to run a simulation:
1. player’s race and AI information
2. initial strength and distribution of both sides
3. map information.
In order to improve the simulation speed and reduce the interference factor, I
decided to plan the above information as follows:
1. player race and AI information:
Test data should include war between three races and ethnic civil wars, a total of six
types of combat.
In order to avoid the simulation error caused by the different types of AI, the AI of
all the battles here adopts the same AI of the nearest attack principle.
Only consider 1v1, the rest of the situation will not be considered.
2. Initial strength and distribution of both sides:
Limit the initial strength of both sides to no more than 100.
The number of each type of arms on both sides is random. For example, if there are
ten types of arms in the Terran, the number of each type of arms is completely random,
but the sum of the ten types of arms should be 100. The position of each unit is
randomly distributed.
3. map information:
In order to avoid interference from other factors, a flat map of 1280 Ã 720 is used here,
and the two armies are distributed on the left and right sides [10].
After the above training set is completed, it is also necessary to place one of them in
the high ground to fight and collect the battle results under the same conditions.
4.2 Training Result
We compared this method with other standard classifiers and divided the results into
two groups for comparison. One group was regarded as the control group in which
both sides were placed on the same terrain, and the other group was regarded as the
experimental group where one of the two sides were placed on unfavorable terrain.
Each group recorded a result with a sample size of 50, 100, 200, 500, and 800 in the
test set. The results are shown in Fig. 2.
128
C. Meng et al.
4.1 Data set Acquisition
In order to develop the speedometer, we will only consider the Army of each party for
the time being, regardless of the air force situation of the three races, and then consider
expanding the Air Force part after the follow-up algorithm is implemented and verified
[4, 9].
SparCraft requires the following information to run a simulation:
1. player’s race and AI information
2. initial strength and distribution of both sides
3. map information.
In order to improve the simulation speed and reduce the interference factor, I
decided to plan the above information as follows:
1. player race and AI information:
Test data should include war between three races and ethnic civil wars, a total of six
types of combat.
In order to avoid the simulation error caused by the different types of AI, the AI of
all the battles here adopts the same AI of the nearest attack principle.
Only consider 1v1, the rest of the situation will not be considered.
2. Initial strength and distribution of both sides:
Limit the initial strength of both sides to no more than 100.
The number of each type of arms on both sides is random. For example, if there are
ten types of arms in the Terran, the number of each type of arms is completely random,
but the sum of the ten types of arms should be 100. The position of each unit is
randomly distributed.
3. map information:
In order to avoid interference from other factors, a flat map of 1280 Ã 720 is used here,
and the two armies are distributed on the left and right sides [10].
After the above training set is completed, it is also necessary to place one of them in
the high ground to fight and collect the battle results under the same conditions.
4.2 Training Result
We compared this method with other standard classifiers and divided the results into
two groups for comparison. One group was regarded as the control group in which
both sides were placed on the same terrain, and the other group was regarded as the
experimental group where one of the two sides were placed on unfavorable terrain.
Each group recorded a result with a sample size of 50, 100, 200, 500, and 800 in the
test set. The results are shown in Fig. 2.
128
C. Meng et al.