sage.architecture.frontend.mscnn1d

Filename = Foobar.py Description = Lorem ipsum dolor sit amet

Created on Tue Nov 21 17:19:53 2021

__author__ = Narenraju Nagarajan __copyright__ = Copyright 2021, Sage __credits__ = nnarenraju __license__ = MIT Licence __version__ = 0.0.1 __maintainer__ = Narenraju Nagarajan __email__ = nagarajan@uni-potsdam.de __status__ = [‘inProgress’, ‘Archived’, inUsage, ‘Debugging’]

Github Repository: NULL

Documentation

Classes

Conv1dSame	1D convolution with padding='same' semantics.
ConcatBlockConv5	Multi-scale inception-style block with five parallel convolutions.
ConvBlock	Three-stage multi-scale 1D CNN frontend for a single detector channel.

Module Contents

class Conv1dSame(in_channels, out_channels, kernel_size, stride=1, dilation=1, groups=1, bias=True)

Bases: torch.nn.Conv1d

1D convolution with padding='same' semantics.

Thin wrapper around torch.nn.Conv1d that always requests padding='same', ensuring the output length matches the input length regardless of kernel size or dilation.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

class ConcatBlockConv5(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True, act=nn.SiLU)

Bases: torch.nn.Module

Multi-scale inception-style block with five parallel convolutions.

Applies five 1D convolutions in parallel at scales k, 2k, k//2, k//4, and 4k, concatenates their outputs together with the identity skip connection, then fuses with a 1×1 pointwise convolution. This allows each block to capture temporal features across a wide range of time scales simultaneously — critical for gravitational-wave signals whose frequency sweeps from low to high over hundreds of milliseconds.

Parameters:

• in_channels (int) – Number of input channels.

• out_channels (int) – Number of output channels per parallel branch (and the final output).

• kernel_size (int) – Base kernel size (k). The five branches use k, 2k, k//2, k//4, and 4k.

• stride (int) – Stride for all parallel convolutions (default 1).

• act (callable) – Activation class to instantiate (default nn.SiLU).

• state (Initialize internal Module)

• ScriptModule. (shared by both nn.Module and)

c1

c2

c3

c4

c5

c6

forward(x)

class ConvBlock(filters_start=32, kernel_start=64, in_channels=1)

Bases: torch.nn.Module

Three-stage multi-scale 1D CNN frontend for a single detector channel.

Processes one detector’s time-domain waveform through three successive ConcatBlockConv5 pairs with decreasing kernel scales and interleaved downsampling:

• Stage 1: kernels k, k//2+1 → MaxPool1d(8) (8× downsample)

• Stage 2: kernels k//2+1, k//4+1 → MaxPool1d(4) (4× downsample)

• Stage 3: kernels k//4+1, k//4+1 (no spatial downsample)

Output is unsqueezed to add a height dimension, converting the 1D time-series into a 2D feature map (B, 1, C, T_down) suitable for the 2D ResNet backend.

Parameters:

• filters_start (int) – Base number of output filters (default 32).

• kernel_start (int) – Base kernel size k (default 64).

• in_channels (int) – Number of input channels (1 for single-detector input, default 1).

• state (Initialize internal Module)

• ScriptModule. (shared by both nn.Module and)

conv1

conv2

conv3

forward(x)