Chapter 27. Learning to Assess Weapons

Key Topics

• Four Different Approaches

• Rationale

• Module Design

• Implementation

• Application

• Evaluation

Now that we've investigated different varieties of decision trees (DTs), we can use them to improve the weapon-selection behaviors. The main problem with the voting system is that it was tedious to set up and used theoretical assumptions to derive the weapon choice.

In this chapter, we focus the power of DT learning algorithms to resolve these problems. Notably, we'll be able to learn about the weapons by experience during the game, which prevents making unfounded assumptions. The learning may also reduce the effort needed to get a working behavior. Because we already have a voting system, however, we could take advantage of it.

This chapter covers the following topics:

• A description of the four different ways we can use the DTs to learn weapon selection

• An analysis of the options and how to pick out the most suitable one, which actually assesses weapons as a whole—as hinted by the chapter's title

• The design of some interfaces to interact with the DT as a modular component

• The implementation of the algorithms discussed in the previous chapter

• The application phase, applying the DT to weapon selection

• A swift evaluation of the system and potential improvements

At the end of the chapter, we'll have a fully working deathmatch bot that's not only capable of moving and shooting, but capable of making tactical decisions about what weapon to use.