reference/generated/tensordict.nn.distributions.TruncatedNormal

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

TruncatedNormal

class tensordict.nn.distributions.TruncatedNormal(loc: Number | Tensor, scale: Number | Tensor, a: Number | Tensor, b: Number | Tensor, validate_args: bool | None = None)

Bases: TruncatedStandardNormal

A normal distribution of mean loc and standard deviation scale, restricted to [a, b].

Its samples are reparameterized (rsample).

Source: https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf

Parameters:

• loc (Number or Tensor) – the mean of the normal distribution before truncation.

• scale (Number or Tensor) – the standard deviation of the normal distribution before truncation. Values under 1e-6 are set to 1e-6.

• a (Number or Tensor) – the lower bound of the support, in the same units as loc.

• b (Number or Tensor) – the upper bound of the support, in the same units as loc.

• validate_args (bool, optional) – passed to Distribution. Defaults to None.

Raises:

ValueError – if a is not smaller than b everywhere.

Examples

>>> import torch
>>> from tensordict.nn.distributions import TruncatedNormal
>>> _ = torch.manual_seed(0)
>>> dist = TruncatedNormal(loc=torch.zeros(3), scale=torch.ones(3), a=-1.0, b=1.0)
>>> dist.sample()
tensor([-0.0064,  0.4764, -0.7754])
>>> dist.mean
tensor([0., 0., 0.])
>>> TruncatedNormal(loc=5.0, scale=1.0, a=4.0, b=6.0).mean
tensor(5.)

cdf(value: Tensor) → Tensor

Returns the cumulative density/mass function evaluated at value.

Parameters:

value (Tensor) –

icdf(value: Tensor) → Tensor

Returns the inverse cumulative density/mass function evaluated at value.

Parameters:

value (Tensor) –

log_prob(value: Tensor) → Tensor

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

Parameters:

value (Tensor) –

property support: Constraint

Returns a Constraint object representing this distribution’s support.