One dimensional convolution is usually used to process text, so the input is usually a long piece of text, which is a list of words

 

The function is defined as follows:

tf.layers.conv1d(
inputs,
filters,
kernel_size,
strides=1,
padding='valid',
data_format='channels_last',
dilation_rate=1,
activation=None,
use_bias=True,
kernel_initializer=None,
bias_initializer=tf.zeros_initializer(),
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
trainable=True,
name=None,
reuse=None
)

The more important parameters are inputs, filters and kernel_ Size, which are described below

 

Inputs: input tensor. Dimension (none, a, b) is a three-dimensional tensor

None: usually the number of samples to be filled, batch_ size

A: the number of words or words in a sentence

B: vector dimension of word or word

 

Filters: the number of filters

 

kernel_ Size: the size of convolution kernel. Convolution kernel should be two-dimensional. Only one dimension needs to be specified here, because the second dimension of convolution kernel is consistent with the input word vector dimension, because for sentences, convolution can only move along the word direction, that is, it can only move in the column dimension

 

An example:

inputs = tf.placeholder (‘float’, shape=[None, 6, 8])

out = tf.layers.conv1d (inputs, 5, 3)

 

Note: for a sample, the sentence length is 6 words, and the dimension of the word vector is 8

filters=5, kernel_ Size = 3, so the dimension of convolution kernel is 3 * 8

Then the input 6 * 8 is convoluted by 3 * 8 convolution kernel to get a vector of 4 * 1 (4 = 6-3 + 1)

And because there are five filters, we get five 4 * 1 vectors

The drawing is as follows: