Environment

MahjongEnv

Multi-agent Mahjong environment where 4 agents play against each other.

options:
  members:
    - reset
    - step
    - get_obs
    - get_oracle_obs
    - get_full_obs
    - get_valid_actions
    - get_payoffs
    - is_over
    - get_curr_player_id
    - render
  show-inheritance:

Attributes

Attribute	Type	Description
observation_space	Box	Executor observation space (93, 34)
oracle_observation_space	Box	Oracle observation space (18, 34)
full_observation_space	Box	Full observation space (111, 34)
action_space	Discrete	Action space (54 actions)

Constants

Constant	Value	Description
PLAYER_OBS_DIM	93	Executor observation channels
ORACLE_OBS_DIM	18	Oracle observation channels
ACTION_DIM	54	Number of discrete actions
MAHJONG_TILE_TYPES	34	Number of tile types

SingleAgentMahjongEnv

Single-agent environment compatible with OpenAI Gym interface.

options:
  members:
    - reset
    - step
    - get_obs
    - get_oracle_obs
    - get_full_obs
    - get_valid_actions
    - render
  show-inheritance:

Parameters

Parameter	Type	Description
opponent_agent	str	Either “random” or path to pretrained model

Attributes

Same as MahjongEnv.