ray.rllib.utils.tf_utils.get_placeholder#

ray.rllib.utils.tf_utils.get_placeholder(*, space: gymnasium.Space | None = None, value: Any | None = None, name: str | None = None, time_axis: bool = False, flatten: bool = True) → tensorflow.compat.v1.placeholder[source]#

Returns a tf1.placeholder object given optional hints, such as a space.

Note that the returned placeholder will always have a leading batch dimension (None).

Parameters:

  • space – An optional gym.Space to hint the shape and dtype of the placeholder.

  • value – An optional value to hint the shape and dtype of the placeholder.

  • name – An optional name for the placeholder.

  • time_axis – Whether the placeholder should also receive a time dimension (None).

  • flatten – Whether to flatten the given space into a plain Box space and then create the placeholder from the resulting space.

Returns:

The tf1 placeholder.