Tag Archives: Tensorflow usage

tf.contrib.layers .xavier_ Initializer function usage

First of all tf.contrib.layers

This is a packaged high-level Library in tensorflow1. X, in which there are many high-level packages of functions,

Convolution function tf.contrib.layers . conv2d(), pooling function tf.contrib.layers .max_ Pool2d() and tf.contrib.layers .avg_ Pool2d (), all join function tf.contrib.layers .fully_ Connected () and so on

Using this high-level library to develop programs will improve efficiency.

Here is an introduction tf.contrib.layers .xavier_ Initializer function

xavier_initializer(
    uniform=True,
    seed=None,
    dtype=tf.float32
)

This function returns an initializer “Xavier” for initializing weights.

This initializer is used to make the variance of each layer output as equal as possible.

Parameters:

Uniform: use uniform or normal distribution to initialize randomly.
seed: can be regarded as seed used to generate random numbers
dtype: only supports floating-point numbers.

Return value:

Initializing the weight matrix
is necessary

Tensorflow2. X does not use contrib advanced library

solution: how to initialize weights by Xavier rules in tensorflow 2.0?