Action Space

Overview

The action space consists of 54 discrete actions. Not all actions are valid at each step - use get_valid_actions() to check.

Action Indices

Discard Actions (0-36)

Index	Action	Description
0-33	Discard	Discard tile type 0-33
34	Discard Red 5m	Discard red 5 Man
35	Discard Red 5p	Discard red 5 Pin
36	Discard Red 5s	Discard red 5 Sou

Chi Actions (37-42)

Index	Action	Description
37	Chi Left	Chi with tiles tile+1, tile+2
38	Chi Middle	Chi with tiles tile-1, tile+1
39	Chi Right	Chi with tiles tile-2, tile-1
40	Chi Left (Red)	Chi Left using red dora
41	Chi Middle (Red)	Chi Middle using red dora
42	Chi Right (Red)	Chi Right using red dora

Pon Actions (43-44)

Index	Action	Description
43	Pon	Pon without using red dora
44	Pon (Red)	Pon using red dora

Kan Actions (45-47)

Index	Action	Description
45	AnKan	Concealed kan
46	MinKan	Open kan (daiminkan)
47	KaKan	Added kan (kakan)

Special Actions (48-53)

Index	Action	Description
48	Riichi	Declare riichi
49	Ron	Win by ron
50	Tsumo	Win by tsumo
51	Push	Kyushukyuhai (nine terminals)
52	Pass Riichi	Pass on riichi decision
53	Pass Response	Pass on response

Riichi Mechanism

Riichi is a two-step action:

1. First step: Discard a tile that can be riichi-declared

2. Second step: Choose between Riichi (48) or Pass Riichi (52)

# Example riichi flow
if 48 in valid_actions:
    # Player can declare riichi
    # First, discard a riichi tile
    riichi_tiles = pm.encv1_get_riichi_tiles(env.t)
    action = riichi_tiles[0]  # Discard first available riichi tile
    env.step(player_id, action)

    # Second step: declare riichi or pass
    valid_actions = env.get_valid_actions()
    # valid_actions will be [48, 52]
    env.step(player_id, 48)  # Declare riichi

Getting Valid Actions

# Get list of valid action indices
valid_actions = env.get_valid_actions()
# Example: [0, 1, 2, 5, 6, 7, 48]

# Get one-hot mask
mask = env.get_valid_actions(nhot=True)
# Shape: (54,), 1 for valid actions, 0 for invalid

Action Constants

The environment provides action constants for convenience:

from pymahjong.env_pymahjong import MahjongEnv

env = MahjongEnv()

# Action indices
env.CHILEFT      # 37
env.CHIMIDDLE    # 38
env.CHIRIGHT     # 39
env.PON          # 43
env.ANKAN        # 45
env.MINKAN       # 46
env.KAKAN        # 47
env.RIICHI       # 48
env.RON          # 49
env.TSUMO        # 50
env.PUSH         # 51
env.PASS_RIICHI  # 52
env.PASS_RESPONSE # 53