LocallyConnected2d

class radionets.architecture.LocallyConnected2d(in_channels, out_channels, output_size, kernel_size, stride, bias=False)

Bases: Module

A 2D locally connected layer implementation.

Unlike convolutional layers that share weights across spatial locations, locally connected layers use different weights for each spatial position. This allows the layer to learn location-specific features while maintaining the sliding window approach of convolutions.

Parameters:

in_channelsint

Number of input channels.

out_channelsint

Number of output channels.

output_sizetuple of int

Expected output spatial dimensions as (height, width).

kernel_sizeint

Size of the sliding window (assumes square kernel).

strideint

Stride of the sliding window (assumes same stride for both dimensions).

biasbool, optional

If True, adds a learnable bias parameter. Default is False.

Attributes:

weightnn.Parameter

Learnable weights with shape (1, out_channels, in_channels, output_height, output_width, kernel_size²).

biasnn.Parameter or None

Learnable bias with shape (1, out_channels, output_height, output_width) if bias=True, else None.

kernel_sizetuple of int

Kernel size as (height, width).

stridetuple of int

Stride as (height, width).

Methods Summary

forward(x)

Define the computation performed at every call.

Methods Documentation

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.