Getting Started

Prerequisites

• Installed pz-battlesnake (see Installation)

Initializating an Enviorment

We can create a new enviorment by importing one of the available enviorments, for simplicity we will use the solo_v0 enviorment.

# First we need to import the enviorment
from pz_battlesnake.env import solo_v0

# Then we can create a new enviorment
env = solo_v0.env() 

the solo_v0 enviorment has parameters which can be passed to the enviorment, for example:

# Create a 15x15 solo enviorment 
env = solo_v0.env(width=15, height=15)

There are more parameters available, see the solo_v0 enviorment for more information.

Interacting with the Enviorment

We now have an enviorment, we can interact with. We need to reset the enviorment before we can interact with it.

env.reset()

Once the enviorment is reset, we can iterate over the agents in the enviorment by looping over all env.agents

# Iterate over the agents in the enviorment
for agent in env.agents:
    # Get Last Observation, Reward, Done, Info
    observation, reward, done, info = env.last()
    # Pick a random action, if agent is done pick None
    action = env.action_space(agent).sample() if not done else None
    # Step through the enviorment
    env.step(action)

Danger

if an agent is done, it needs to be None, otherwise the enviorment will throw an error.

Running Multiple Games

The current code only runs through 1 game, to run through more than 1 game, we need to reset the enviorment after each game.

# Run for 10 games
for _ in range(10):
    # Reset enviorment, before each game
    env.reset() 
    done = False
    while not done:
        for agent in env.agents:
            # Get Last Observation, Reward, Done, Info
            observation, reward, done, info = env.last()
            # Pick a random action, if agent is done pick None
            action = env.action_space(agent).sample() if not done else None
            # Step through the enviorment
            env.step(action)
        # This is a shortcut to set the done to be true, 
        # since when all agents are done the env.agents array will be empty
        done = not env.agents

Useful links

• PettingZoo API Docs

Warning

Some of the things listed on PettingZoo’s Guide are not available in pz-battlesnake, as we implement an AEC Enviorment and not an MARL Enviorment. Below is listed what you can’t do. - PettingZoo Parrallel API - Raw Enviorments

Complete Code

Note

The code below implements a random agent, which will take a random action each step.

# First we need to import the enviorment
from pz_battlesnake.env import solo_v0

# Then we can create a new enviorment
env = solo_v0.env() 

# Create a 15x15 solo enviorment 
# env = solo_v0.env(width=15, height=15) # uncomment this to create a 15x15 solo enviorment

# Run for 10 games
for _ in range(10):
    # Reset enviorment, before each game
    env.reset()
    done = False
    while not done:
        for agent in env.agents:
            # Get Last Observation, Reward, Done, Info
            observation, reward, done, info = env.last()
            # Pick an action, if agenmt is done, pick None
            action = env.action_space(agent).sample() if not done else None
            # Step through the enviorment
            env.step(action)
        # Code below runs, when all agents has taken an action
        # Render the enviorment
        # env.render() # uncomment this to render
        # This is a shortcut to set the done to be true, 
        # since when all agents are done the env.agents array will be empty
        done = not env.agents