OpenSpiel game methods: policytensorshape

Back to Core API reference

policy_tensor_shape()

Returns the shape of the policy tensor as a list of integers. This describes the structure of a policy vector over actions. By default this is [num_distinct_actions()], representing a flat probability distribution over all actions. Games may override this to provide a structured shape (e.g. a 2D grid for board games).

Examples:

import pyspiel

game = pyspiel.load_game("tic_tac_toe")
print(game.policy_tensor_shape())        # Output: [9]
print(game.num_distinct_actions())       # Output: 9

game = pyspiel.load_game("leduc_poker")
print(game.policy_tensor_shape())        # Output: [4]