Base class

[source code]

AgentBase

connectFourLab.game.agents.basicAgents.AgentBase()

Base class Agent

To inherit from this class it's required to implement the action method

def action(self, board):
    # choose a column based in a given board
    return chosen_column # 0-6

Attributes

• name - str, required, name of the agent
• description - str, required, description of the agent
• kind - str, required, category of the agent
• model_key - str, required for agents who uses neural network models
• clock_management - bool, optional, default False (*)
  • flag that indicates whether or not the agent can manage a time limit game
• require_nn_model - bool, optional, default False (*)
  • flag that indicates whether or not the agent needs a neural network model

(*) - only necessary in the app interface

Saving the data

You can easily save the data from all turns of the game using save all data will be saved separately in:

• data_scenario - state of the board
• data_action - taken action
• data_reward - attributed reward

def action(self, board):
    choice = self.random_choice(board)
    self.save(board, choice)
    return choice

Exception

• BadImplementation - some class didn't implemented the required method(s)

Clock management

If the game is time limited the update_clock will be called every turn and you will be able to manage the Timer (see Timer) in self.clock.

Using Neural Networks

To use neural network you need a Trainer which have to create and train a model for the agent.

The trained models are stored in the game/models/ folder. Example

• MCTSNN - Monte Carlo Tree Search with Neural Network implementation - see documentation