dropout1d

→Tensor

dropout1d(x: Tensor, p: float = 0.5, training: bool = True)

source

Channel-wise dropout for 1-D sequence / feature inputs.

Drops entire 1-D feature maps (channels) of an $(N, C, L)$ tensor with probability p, scaling the survivors by $1/(1-p)$ to keep the activation magnitude unbiased. Standard per-element dropout is statistically weak when applied within strongly-correlated feature maps (consecutive positions in a sequence are highly correlated), so masking the whole channel at once provides stronger regularisation in 1-D CNNs and temporal feature extractors.

Parameters

xTensor

Input tensor, typically of shape

(N, C, L)

pfloat= 0.5

Channel-drop probability in

[0, 1)

(default 0.5).

trainingbool= True

When False, returns x unchanged — identity at inference time (default True).

Returns

Tensor

Same shape and dtype as x.

Notes

For each batch element $n$ and channel $c$ , draw $m_{n,c} \sim \mathrm{Bernoulli}(1 - p)$ independently and apply

y_{n, c, \ell} = \frac{m_{n, c}}{1 - p} \cdot x_{n, c, \ell}.

The $1/(1-p)$ scale ("inverted dropout") preserves $\mathbb{E}[y] = \mathbb{E}[x]$ so no rescaling is needed at inference.

Examples

>>> import lucid
>>> from lucid.nn.functional import dropout1d
>>> x = lucid.ones((2, 4, 5))
>>> y = dropout1d(x, p=0.5, training=True)