The GameAgent Plugin - Martyn0324/SerpentAI GitHub Wiki

The GameAgent plugin is meant to represent a specific extension of the base GameAgent class.

• A plugin definition file
• A GameAgent subclass
• Bundled helper modules
• Bundled machine learning models

The GameAgent subclass is the most important file of your project. It contains your implementations (frame handlers) that will define the logic used to tackle the target game. Here is what is to be provided:

• One or more frame handlers functions
• Zero or more frame handler setup functions
• Frame handler registrations inside self.frame_handlers in the constructor
• Frame handler setup registrations inside self.frame_handler_setups in the constructor
• An optional instance of AnalyticsClient in self.analytics_client in the constructor

You are free to implement more custom instance methods. If the size of the file becomes unmanageable (it happens; game agents are pretty complex!), start using extracting functions to helper modules.

1. Add an instance method to the GameAgent subclass for your frame handler
2. Add a dictionary entry in self.frame_handlers in the constructor, mapping a name to the function.

class MyGameAgent(GameAgent):
    def __init__(self, **kwargs):
        super().init(**kwargs)

        self.frame_handlers["MY_FRAME_HANDLER"] = self.my_frame_handler

    def my_frame_handler(self, game_frame):
        pass

Exactly like Frame Handlers but in self.frame_handler_setups. They need to share the same name as their Frame Handler counterparts.

class MyGameAgent(GameAgent):
    def __init__(self, **kwargs):
        super().init(**kwargs)

        self.frame_handler_setups["MY_FRAME_HANDLER"]: self.my_frame_handler_setup

    def my_frame_handler_setup(self):
        pass

1. Create new files in files/helpers
2. Populate them with the desired functions and classes
3. Import them in your GameAgent subclass. These loose files are not made available by the plugin system so a relative import is needed: from .helpers.<my_module> import *

In files/helpers/utils.py_ a hello_world function is defined:

def hello_world():
    print("Hello World!")

To access it in your GameAgent subclass, you add this import statement:

from .helpers.utils import hello_world

Simply copy them to files/ml_models. Renaming them to the hold the .model extension is recommended because when you make a Git repository for your plugin, .model files are added using Git LFS which enables uploads of files over 100MB.

Simply add a dictionary entry in self.machine_learning_models, mapping a name to the result of self.load_machine_learning_model.

To make a machine learning model available to all your frame handlers, load it in the GameAgent's constructor:

class MyGameAgent(GameAgent):
    def __init__(self, **kwargs):
        super().init(**kwargs)

        self.machine_learning_models["MY_MODEL"] = self.load_machine_learning_model(
            os.path.join(os.path.dirname(__file__), "ml_models/my_model.model")
        )

To limit a machine learning model to only be in scope for a single frame handler, load it in the latter's setup function:

class MyGameAgent(GameAgent):
    def __init__(self, **kwargs):
        super().init(**kwargs)

        self.frame_handlers["MY_FRAME_HANDLER"]: self.my_frame_handler
        self.frame_handler_setups["MY_FRAME_HANDLER"]: self.my_frame_handler_setup

    def my_frame_handler_setup(self):
        self.machine_learning_models["MY_MODEL"] = self.load_machine_learning_model(
            os.path.join(os.path.dirname(__file__), "ml_models/my_model.model")
        )

    def my_frame_handler(self, game_frame):
        pass