Now that you have decided to use deep learning and/or deep reinforcement learning for your game, it is time to determine how you plan to implement various functionality in your game. In order to do that, we are going to go through a table outlining the steps you need to go through in order to build your game's AI agent:
|
Step |
Action |
Summary |
| Start | Determine at what level you want the AI in the game to operate, from basic, perhaps for just testing and simple automation, to advanced, where the AI will complete against the player. | Determine the level of AI. |
| Resourcing | Determine the amount of resources. Basic AI or automation could be handled within the team itself, whereas more complex AI may require one or many experienced members of staff. | Team requirements. |
| Knowledge | Determine the level of knowledge the team possesses and what will be required. It is a given that any team implementing new AI will need to learn new skills. | Knowledge-gap analysis. |
| Demonstration | Always start by building a simple but workable proof of concept that demonstrates all critical aspects of the system. | Demonstrate the team can complete the basic premise. |
| Implementation | Build the actual system in a way that is simplistic and maintainable. Keep all the things you know simple and clean. | Build the system. |
| Testing | Test the system over and over again. It is critical that the system is tested thoroughly, and of course what better way to do that than with a DRL automated test system. | Test the system. |
| Fix | As anyone who has developed software for more than a few weeks will tell you, the process is build, test, fix, and repeat. That essentially is the software development process, so try not to add too many other bells and whistles to distract from that. | Fixing the system. |
| Release | Releasing software to users/players is absolutely critical to a successful game or software product of any kind. You will always want to release early and often, which means your players must be encouraged to test, and to provide feedback. | Let the bugs out. |
| Repeat | The cycle is endless and will continue as long as your product/game makes money. | Support the system. |
The preceding process is the basic premise and will work for most of your development needs. In most cases, you may want to track individual work items such as features or bugs on a work or task board. You may want to use a more defined process such as Scrum, but often keeping things simple is your best course of action.
Scrum and other software development processes are great examples to learn from, but unless you have formally trained staff, it's better to avoid trying to implement these yourself. There are often subtle rules that need to be enforced in these processes for them to work as they claim to. Even trained Scrum Masters may need to battle daily to enforce these rules in many organizations, and in the end their value becomes more management-driven than developer-focused. Use the previous table as a guide for the steps you take in building your next game, and always remember that build, release, fix, and repeat is the key to good software.
In the next section, we will look at other things you can use to expand your learning.