reference/generated/tensordict.nn.distributions.Delta

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

Delta

class tensordict.nn.distributions.Delta(param: Tensor, atol: float = 1e-06, rtol: float = 1e-06, batch_shape: Sequence[int] | Size | None = None, event_shape: Sequence[int] | Size | None = None)

Bases: Distribution

A distribution with all its mass on one value, param.

sample() and rsample() return param (expanded to the sample shape), and mean, mode and deterministic_sample return it as it is. log_prob() returns inf for a value equal to param (within atol and rtol, on every element of the event) and -inf otherwise. It is the default distribution_class of ProbabilisticTensorDictModule, which then copies its input to its output.

Parameters:

• param (Tensor) – the value that holds all the mass.

• atol (float, optional) – the absolute tolerance with which log_prob() compares a value with param. Defaults to 1e-6.

• rtol (float, optional) – the relative tolerance with which log_prob() compares a value with param. Defaults to 1e-6.

• batch_shape (Size, optional) – the batch shape. Defaults to None: see event_shape.

• event_shape (Size, optional) – the shape of one value. If neither batch_shape nor event_shape is given (or both are empty), the last dimension of param is the event shape and the others are the batch shape. Defaults to None.

Examples

>>> import torch
>>> from tensordict.nn.distributions import Delta
>>> dist = Delta(torch.zeros(4, 2))
>>> dist.batch_shape, dist.event_shape
(torch.Size([4]), torch.Size([2]))
>>> dist = Delta(torch.tensor([1.0, 2.0]))
>>> dist.sample()
tensor([1., 2.])
>>> dist.log_prob(torch.tensor([1.0, 2.0])), dist.log_prob(torch.tensor([1.0, 3.0]))
(tensor(inf), tensor(-inf))

log_prob(value: Tensor) → Tensor

Returns the log of the probability density/mass function evaluated at value.

Parameters:

value (Tensor) –

property mean: Tensor

Returns the mean of the distribution.

property mode: Tensor

Returns the mode of the distribution.

rsample(sample_shape: Sequence[int] | Size | None = None) → Tensor

Generates a sample_shape shaped reparameterized sample or sample_shape shaped batch of reparameterized samples if the distribution parameters are batched.

sample(sample_shape: Sequence[int] | Size | None = None) → Tensor

Generates a sample_shape shaped sample or sample_shape shaped batch of samples if the distribution parameters are batched.

update(param: Tensor) → None

Replaces param. The batch and event shapes do not change.