Skip to content

Latest commit

 

History

History
111 lines (84 loc) · 4.29 KB

README.md

File metadata and controls

111 lines (84 loc) · 4.29 KB
Raw

PyLoL OpenAI Gym Environments

OpenAI Gym Environments for the League of Legends v4.20 PyLoL environment.

Installation

You can install LoLGym from a local clone of the git repo:

git clone https://github.com/MiscellaneousStuff/lolgym.git
pip3 install -e lolgym/

Usage

You need the following minimum code to run any LoLGym environment:

Import gym and this package:

import gym
import lolgym.envs

Import and initialize absl.flags (required due to pylol dependency)

import sys
from absl import flags
FLAGS = flags.FLAGS
FLAGS(sys.argv)

Create and initialize the specific environment.

Available Environments

LoL1v1

The full League of Legends v4.20 game environment. Initialize as follows:

env = gym.make("LoLGame-v0")
env.settings["map_name"] = "New Summoners Rift" # Set the map
env.settings["human_observer"] = False # Set to true to run league client
env.settings["host"] = "localhost" # Set this to a local ip
env.settings["players"] = "Nidalee.BLUE,Lucian.PURPLE"

The players setting specifies which champions are in the game and what team they are playing on. The pylol environment expects them to be in a comma-separated list of Champion.TEAM items with that exact capitalization.

Versions:

  • LoLGame-v0: The full game with complete access to action and observation space.

LoL1DEscape

Minigame where the controlling agent must maximize it's distance from the other agent by moving either left or right. Initialize as follows:

env = gym.make("LoL1DEscape-v0")
env.settings["map_name"] = "New Summoners Rift" # Set the map
env.settings["human_observer"] = False # Set to true to run league client
env.settings["host"] = "localhost" # Set this to a local ip
env.settings["players"] = "Nidalee.BLUE,Lucian.PURPLE"

Versions:

  • LoL1DEscape-v0: Highly stripped version of LoL1v1 where the only observation is the controlling agents distance from the enemy agent and the only action is to move left or right.

Notes

  • The action space for this environment doesn't require the call to functionCall like pylol does. You only need to call it with an array of action and arguments. For example:

      _SPELL = actions.FUNCTIONS.spell.id
  _EZREAL_Q = [0]
  _TARGET = point.Point(8000, 8000)
  acts = [[_SPELL, _EZREAL_Q, _TARGET] for _ in range(env.n_agents)]
  obs_n, reward_n, done_n, _ = env.step(acts)

    The environment will not check whether an action is valid before passing it along to the pysc2 environment so make sure you've checked what actions are available from obs.observation["available_actions"].

  • This environment doesn't specify the observation_space and action_space members like traditional gym environments. Instead, it provides access to the observation_spec and action_spec objects from the pylol environment.

General Notes

  • Per the Gym environment specifications, the reset function returns an observation, and the step function returns a tuple (observation_n, reward_n, done_n, info_n), where info_n is a list of empty dictionaries. However, because lolgym is a multi-agent environment each item is a list of items, i.e. observation_n is an observation for each agent, reward_n is the reward for each agent, done_n is whether any of the observation.step_type is LAST.
  • Aside from step() and reset(), the environments define a save_replay() method, that accepts a single parameter replay_dir, which is the name of the replay directory to save the GameServer replays inside of.
  • All the environments have the following additional properties:
    • episode: The current episode number
    • num_step: The total number of steps taken
    • episode_reward: The total reward received for this episode
    • total_reward: The total reward received for all episodes
  • The examples folder contains examples of using the various environments.