reference/generated/torchrl.objectives.two_hot_cross_entropy

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

torchrl.objectives.two_hot_cross_entropy

torchrl.objectives.two_hot_cross_entropy(logits: Tensor, target: Tensor, bins: Tensor) → Tensor

Return two-hot cross entropy for raw scalar targets.

Parameters:

• logits (torch.Tensor) – Categorical logits with bins in the trailing dimension.

• target (torch.Tensor) – Raw scalar targets, optionally with a trailing singleton dimension.

• bins (torch.Tensor) – One-dimensional support in raw value space.

Returns:

The unreduced cross entropy with the trailing category dimension removed.

Examples

>>> import torch
>>> from torchrl.objectives import two_hot_cross_entropy
>>> logits = torch.zeros(2, 3)
>>> target = torch.tensor([-0.5, 0.5])
>>> two_hot_cross_entropy(logits, target, torch.tensor([-1.0, 0.0, 1.0]))
tensor([1.0986, 1.0986])