TruncatedNormalNoise

TruncatedNormalNoise(
   mu: Union[float, th.Tensor] = 0.0, sigma: Union[float, th.Tensor] = 1.0,
   low: float = -1.0, high: float = 1.0, eps: float = 1e-06,
   stddev_schedule: str = 'linear(1.0, 0.1, 100000)'
)

Truncated normal action noise. See Section 3.1 of "Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning".

Args

mu (Union[float, th.Tensor]) : Mean of the noise.
sigma (Union[float, th.Tensor]) : Standard deviation of the noise.
low (float) : The lower bound of the noise.
high (float) : The upper bound of the noise.
eps (float) : A small value to avoid numerical instability.
stddev_schedule (str) : Use the exploration std schedule, available options are: linear(init, final, duration) and step_linear(init, final1, duration1, final2, duration2).

Returns

Truncated normal noise instance.

Methods:

.sample

source

.sample(
   clip: Optional[float] = None, sample_shape: th.Size = th.Size()
)

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

Args

clip (Optional[float]) : The clip range of the sampled noises.
sample_shape (th.Size) : The size of the sample to be drawn.

Returns

A sample_shape shaped sample.

.mean

source

.mean()

Returns the mean of the distribution.

.mode

source

.mode()

Returns the mode of the distribution.