2 Related Technology Introduction
2.1 Maximum Posterior Probability Estimate [2, 3]
In Bayesian statistics, the “maximum posterior probability estimate” is the mode of the
posterior probability distribution. Point estimates of quantities that are not directly
observable in the experimental data can be obtained using the maximum a posteriori
probability estimate. It is closely related to the classical method of maximum likelihood
estimation, but it uses an augmented optimization goal to further consider the prior
probability distribution of the estimated quantity. Therefore, the maximum posterior
probability estimate can be regarded as the maximum likelihood estimate of
regularization.
Although the use of posteriori distribution was shared by the maximum a posteriori
estimate and the Bayesian statistic, MAP is generally not considered a Bayesian
method, the maximum posterior estimate is a point estimate; whereas, the Bayesian
method is characterized by using these distributions to summarize the data and get
inferences. The Bayesian method attempts to calculate the posterior mean or median
and the posterior interval, not the posterior model. This is especially true when the
posterior distribution does not have a simple analytical form: In this case, the posterior
distribution can be modeled using Markov chain Monte Carlo techniques, but it is
difficult to find the optimization of its modulus.
2.2 StarCraft Combat Simulation System SparCraft
SparCraft is an open-source abstract StarCraft combat simulation package for Windows
and Linux. It can be used to create stand-alone combat simulations or import into
existing BWAPI-based StarCraft robots to provide additional AI functionality.
SparCraft is built very fast and uses a frame fast-forward system to bypass game
frames, during which no combat decisions are made (i.e., during attack cooling,
moving, etc.). With this system, SparCraft can easily perform tens of millions of unit
actions per second.
For training and testing purposes, we generated data sets using the StarCraft combat
simulator (developed by SparCraft, UAlberta). The simulator allows for the creation
and execution of battles based on deterministic scripts (or decision agents such as
confrontational search algorithms). We chose to use simulation data instead of data
from real games, mainly because the simulator allowed us to generate large amounts of
data and avoid unnecessary interference with data results due to different player game
styles and operating techniques [4, 5].
3 Mathematical Modeling
In StarCraft, the victory of a battle depends largely on a series of parameters such as
damage, range of attack, hit rate, armor value, and speed of each unit in both armies.
Therefore, in the model, we need to build a set of characteristic attributes for each unit,
and the system calculates the result based on the attribute values of each unit in an
Battle Prediction System in StarCraft Combined with Topographic …
123
2.1 Maximum Posterior Probability Estimate [2, 3]
In Bayesian statistics, the “maximum posterior probability estimate” is the mode of the
posterior probability distribution. Point estimates of quantities that are not directly
observable in the experimental data can be obtained using the maximum a posteriori
probability estimate. It is closely related to the classical method of maximum likelihood
estimation, but it uses an augmented optimization goal to further consider the prior
probability distribution of the estimated quantity. Therefore, the maximum posterior
probability estimate can be regarded as the maximum likelihood estimate of
regularization.
Although the use of posteriori distribution was shared by the maximum a posteriori
estimate and the Bayesian statistic, MAP is generally not considered a Bayesian
method, the maximum posterior estimate is a point estimate; whereas, the Bayesian
method is characterized by using these distributions to summarize the data and get
inferences. The Bayesian method attempts to calculate the posterior mean or median
and the posterior interval, not the posterior model. This is especially true when the
posterior distribution does not have a simple analytical form: In this case, the posterior
distribution can be modeled using Markov chain Monte Carlo techniques, but it is
difficult to find the optimization of its modulus.
2.2 StarCraft Combat Simulation System SparCraft
SparCraft is an open-source abstract StarCraft combat simulation package for Windows
and Linux. It can be used to create stand-alone combat simulations or import into
existing BWAPI-based StarCraft robots to provide additional AI functionality.
SparCraft is built very fast and uses a frame fast-forward system to bypass game
frames, during which no combat decisions are made (i.e., during attack cooling,
moving, etc.). With this system, SparCraft can easily perform tens of millions of unit
actions per second.
For training and testing purposes, we generated data sets using the StarCraft combat
simulator (developed by SparCraft, UAlberta). The simulator allows for the creation
and execution of battles based on deterministic scripts (or decision agents such as
confrontational search algorithms). We chose to use simulation data instead of data
from real games, mainly because the simulator allowed us to generate large amounts of
data and avoid unnecessary interference with data results due to different player game
styles and operating techniques [4, 5].
3 Mathematical Modeling
In StarCraft, the victory of a battle depends largely on a series of parameters such as
damage, range of attack, hit rate, armor value, and speed of each unit in both armies.
Therefore, in the model, we need to build a set of characteristic attributes for each unit,
and the system calculates the result based on the attribute values of each unit in an
Battle Prediction System in StarCraft Combined with Topographic …
123