TensorFlow · Class

layers.Conv1D

A Keras layer that performs one-dimensional convolution over spatial or temporal sequences. It transforms sliding input windows into a configurable number of output channels.

How it works

Conv1D slides a learned kernel across one spatial or temporal axis. It optionally adds a bias and activation, converting each input window into an output with the configured number of filters.

Syntax

tf.keras.layers.Conv1D(filters, kernel_size, strides=1, padding='valid', data_format=None, dilation_rate=1, groups=1, activation=None, use_bias=True, kernel_initializer='glorot_uniform', bias_initializer='zeros', kernel_regularizer=None, bias_regularizer=None, activity_regularizer=None, kernel_constraint=None, bias_constraint=None, **kwargs)

Parameters

Name	Type	Description
`filters`	`int`	Number of output filters and output channels.
`kernel_size`	`int or tuple/list of 1 integer`	Width of the one-dimensional convolution window.
`strides`	`int or tuple/list of 1 integer`	Step size of the convolution window.
`padding`	`string`	Padding mode: valid, same, or causal.
`data_format`	`string or None`	Input/output channel layout.
`dilation_rate`	`int or tuple/list of 1 integer`	Dilation rate of the convolution.
`groups`	`int`	Number of input channels grouped together for convolution.
`activation`	`string or callable or None`	Optional activation function.
`use_bias`	`bool`	Whether to add a bias vector.
`kernel_initializer`	`initializer`	Initializer for the convolution kernel.
`bias_initializer`	`initializer`	Initializer for the bias vector.
`kernel_regularizer`	`regularizer or None`	Regularizer for the kernel.

Returns

A 3D output tensor. With channels_last it has shape (batch_shape, new_steps, filters); with channels_first it has shape (batch_shape, filters, new_steps).

Key points

Conv1D is designed for one-dimensional spatial or temporal convolution.
A learned kernel moves across the input axis to generate output features.
The layer can add a trainable bias vector to its convolution results.
An optional activation function is applied after the convolution and bias addition.
Padding accepts valid, same, or causal, without case sensitivity.
The filters argument sets how many output feature channels the layer produces.

Common mistakes

Do not combine a stride greater than one with a dilation rate greater than one.
With valid padding, the temporal dimension can shrink because no border padding is added.
The default channels-last input must be rank three, with steps before channels.

Example: Apply a temporal convolution

import tensorflow as tf

layer = tf.keras.layers.Conv1D(
    filters=16, kernel_size=3, padding="same", activation="relu"
)
x = tf.random.normal((4, 10, 8))
y = layer(x)
print(y.shape)  # (4, 10, 16)

Creates a Conv1D layer, applies it to a batch of 10-step sequences with 8 channels, and prints the resulting shape.

When to use it

Extract local patterns across time steps.

Temporal sequence modeling

Learn features from ordered sensor or signal samples.

One-dimensional signal processing

Detect local patterns in embedded token sequences.

Text classification

Use sliding windows to model fixed-width recent history.

Time-series forecasting

Interview questions

QWhat does Conv1D compute and what shapes does it use?

Conv1D applies each learned filter across a single spatial or temporal dimension. For channels-last data, the input is shaped as batch, steps, channels, and the output is batch, new steps, filters.

Follow-up: How would you preserve the number of steps? Use padding="same" when the stride is one; the resulting output length matches the input length.

QWhat are filters and kernel_size responsible for?

filters determines the output-space size and therefore the number of channels in the result. kernel_size determines the width of the sliding convolution window.

Follow-up: What changes when data_format="channels_first"? The input and output place channels before the steps dimension.

QHow do bias, activation, and padding affect a Conv1D layer?

The layer can add a bias vector when use_bias is enabled and then apply activation when activation is not None. Its padding choices are valid, same, and causal.

Follow-up: When is causal padding useful? It prevents an output at a time step from depending on later input steps, making it suitable for some temporal models.

tf.keras.layers.Convolution1D
tf.keras.layers.DepthwiseConv1D
tf.keras.layers.SeparableConv1D
tf.keras.layers.Conv1DTranspose

Source: tf.keras.layers.Conv1D | TensorFlow v2.16.1. Checked 2026-10-15.