ray.rllib.utils.torch_utils.reduce_mean_ignore_inf#

ray.rllib.utils.torch_utils.reduce_mean_ignore_inf(x: numpy.array | jnp.ndarray | tf.Tensor | torch.Tensor, axis: int | None = None) → numpy.array | jnp.ndarray | tf.Tensor | torch.Tensor[source]#

Same as torch.mean() but ignores -inf values.

Parameters:

  • x – The input tensor to reduce mean over.

  • axis – The axis over which to reduce. None for all axes.

Returns:

The mean reduced inputs, ignoring inf values.