OpenSpiel state methods: legal_actions_mask

Back to Core API reference

1. legal_actions_mask(player: int)

2. legal_actions_mask()

Returns a binary vector where 1 indicates a legal action and 0 indicates an illegal action. The length is game.num_distinct_actions() for player nodes, or game.max_chance_outcomes() for chance nodes. The parameterless version uses the current player.

This is useful for masking illegal actions in neural network policy outputs.

Examples:

import pyspiel

game = pyspiel.load_game("tic_tac_toe")
state = game.new_initial_state()
print(state.legal_actions_mask())
# Output: [1, 1, 1, 1, 1, 1, 1, 1, 1]  (all 9 squares available)

state.apply_action(4)   # Player 0 takes center
print(state.legal_actions_mask())
# Output: [1, 1, 1, 1, 0, 1, 1, 1, 1]  (center no longer available)

# Also works with explicit player argument
print(state.legal_actions_mask(1))
# Output: [1, 1, 1, 1, 0, 1, 1, 1, 1]