torch.js has not been released yet.

torch.nn.ReplicationPad1d

class ReplicationPad1d extends Module

new ReplicationPad1d(padding: Padding1D)

readonlypadding([number, number])

1D replication padding: pads by repeating edge values.

Extends sequences by replicating the edge values. Unlike reflection which mirrors, replication simply repeats the boundary value for the entire padding. Simpler than reflection, good for avoiding boundary artifacts when signal naturally "continues". Essential for:

Signal processing (repeated edge assumption)
Time series with constant boundary behavior
Simpler alternative to reflection padding
Performance-critical applications (simpler computation)

Replication padding repeats edge values: for input [a, b, c] with padding=2, left becomes [a, a | a, b, c], right becomes [a, b, c | c, c].

When to use ReplicationPad1d:

Simpler alternative to reflection
Signal assumes edge value continues
Performance matters (faster than reflection)
Avoiding reflection's boundary limit constraints

Trade-offs:

vs ReflectionPad: Replication simpler; reflection smoother
vs ZeroPad: Replication preserves signal; zero adds discontinuity
Computation: Simpler than reflection (just repeat)
Gradient: Straightforward backprop
No limit: Can pad any amount (unlike reflection's width-1 limit)

Simple edge repeat: Edge value replicated for entire padding
No limit: Can pad more than width (unlike reflection)
Smooth signal: Assumes signal continues at boundary
Simpler computation: Less complex than reflection

Edge repetition visible: Creates obvious pattern at boundary
Not smooth: Sharp repetition unlike reflection's gradual mirroring

Examples

// Basic replication
const pad = new torch.nn.ReplicationPad1d(2);
const x = torch.tensor([[[1, 2, 3, 4, 5]]]);
const y = pad.forward(x);
// Input:  [1, 2, 3, 4, 5]
// Output: [1, 1, 1, 2, 3, 4, 5, 5, 5]  (edges repeated)

See Also

PyTorch torch.nn.ReplicationPad1d

RemovableHandle.remove

ReplicationPad2d