converting xavier to glorot in tensorflow code

Tags: , , ,



I am trying to convert xavier initializer in a tensorflow code to glorot initializer.

I am not sure which one of the following (or other options) is correct?

#initializer = tf.contrib.layers.xavier_initializer()
initializer = tf.initializers.glorot_uniform()  
initializer = tf.initializers.GlorotNormal()

Answer

Based on the code for xavier_initializer, the default is uniform=True which suggests glorot_uniform may be what you want.

If you are using tf2, you may use tf.keras.initializers.GlorotUniform(). tf.keras.initializers.glorot_uniform is just a shortcut function for it: https://www.tensorflow.org/api_docs/python/tf/keras/initializers/GlorotUniform



Source: stackoverflow