class

LPPool1d

extendsModule

LPPool1d(norm_type: float, kernel_size: int, stride: int | None = None, ceil_mode: bool = False)

source

Applies 1-D power-average (Lp-norm) pooling over a sequence.

For each window of length kernel_size the module computes the $\ell^p$ norm of the absolute values of the elements:

y[n, c, i] = \left( \sum_{k=0}^{k_s - 1} \left|x[n,\, c,\, i \cdot s + k]\right|^p \right)^{1/p}

where $p$ = norm_type, $k_s$ = kernel_size, and $s$ = stride.

Special cases:

$p = 1$ — sum pooling (L1 norm over the window).
$p = 2$ — square-root of sum of squares (L2 norm).
$p \to \infty$ — approaches max pooling.

Parameters

norm_typefloat

The exponent

p

. Must be positive and finite.

kernel_sizeint

Size of the sliding window.

strideint or None= None

Step between windows. Defaults to kernel_size.

ceil_modebool= False

Use ceiling for output length. Default: False.

Attributes

norm_typefloat

kernel_sizeint

strideint

ceil_modebool

Notes

Input: (N, C, L_in)
Output: (N, C, L_out) where $L_{out} = \lfloor (L_{in} - k_s) / s \rfloor + 1$

Lp pooling is a power-mean generalisation: it smoothly interpolates between sum pooling ( $p=1$ ), energy pooling ( $p=2$ ), and max pooling ( $p \to \infty$ ).
Because the operation is differentiable for finite $p$ , it can be used as a drop-in replacement for max or average pooling in gradient-based training.
Implemented via engine primitives unfold_dim, abs, pow_scalar, and sum — fully GPU-compatible.

Examples

L2 power pooling (energy pooling):
>>> import lucid
>>> import lucid.nn as nn
>>> pool = nn.LPPool1d(norm_type=2, kernel_size=4)
>>> x = lucid.ones((1, 8, 16))
>>> y = pool(x)
>>> y.shape
(1, 8, 4)
Sum pooling (p=1) with overlapping windows:
>>> pool = nn.LPPool1d(norm_type=1, kernel_size=3, stride=1)
>>> x = lucid.ones((2, 4, 12))
>>> y = pool(x)
>>> y.shape
(2, 4, 10)

Methods (3)

dunder

init

→None

__init__(norm_type: float, kernel_size: int, stride: int | None = None, ceil_mode: bool = False)

source

Initialise the LPPool1d module. See the class docstring for parameter semantics.

forward

→Tensor

forward(x: Tensor)

source

Apply the pooling operation to the input tensor.

Parameters

inputTensor

Input tensor of shape

(N, C, *)

where

*

are the spatial dimensions appropriate for this pooling layer.

Returns

Tensor

Pooled output tensor.

extra_repr

→str

extra_repr()

source

Return a string representation of the layer's configuration.