torch.js has not been released yet.

torch.nn.ZeroPad1d

class ZeroPad1d extends Module

new ZeroPad1d(padding: Padding1D)

readonlypadding([number, number])

1D zero padding: pads by filling with zeros.

Simplest padding mode: extends sequences with zeros. Creates discontinuity at boundary (sharp transition from signal to zeros), but is computationally simple. Standard choice for most deep learning tasks when boundary artifacts are acceptable.

When to use ZeroPad1d:

Standard deep learning (CNNs, RNNs)
Tasks not sensitive to boundary artifacts
Default padding choice
Simple, fast implementation
When other padding modes show no improvement

Trade-offs:

vs ReflectionPad: Zero simpler; reflection avoids artifacts
vs ReplicationPad: Zero creates discontinuity; replication preserves signal
vs CircularPad: Zero for open signals; circular for periodic
Computation: Simplest (just zeros)
Gradient: Straightforward backprop, zeros don't contribute

Standard choice: Default padding in most deep learning
No signal preservation: Zeros at boundary regardless of input
Discontinuity: Creates sharp zero transition
Simple: No special computation needed
Gradient: Padding region contributes zeros to gradient

Examples

// Basic zero padding
const pad = new torch.nn.ZeroPad1d(2);
const x = torch.tensor([[[1, 2, 3]]]);
const y = pad.forward(x);
// Input:  [1, 2, 3]
// Output: [0, 0, 1, 2, 3, 0, 0]

See Also

PyTorch torch.nn.ZeroPad1d

UpsamplingNearest2d