OpenSpiel game methods: observation_tensor_layout

Back to Core API reference

observation_tensor_layout()

Returns the tensor layout used for the observation tensor representation. The layout is either TensorLayout.CHW (channels, height, width) or TensorLayout.HWC (height, width, channels). Defaults to CHW for most games.

This is relevant when interpreting the flat vector returned by state.observation_tensor() according to the shape given by game.observation_tensor_shape().

Examples:

import pyspiel

game = pyspiel.load_game("leduc_poker")
print(game.observation_tensor_layout())
# Output: TensorLayout.CHW

print(game.observation_tensor_shape())
# Output: [16]