What are the differences between tf.initialize_all_variables() and tf.global_variables_initializer()

Question

On Tensorflow official website, it gives explantions of the tf.initialize_all_variables() and tf.global_variables_initializer() functions as follow

tf.initialize_all_variables():

Returns an op that initializes all variables.

tf.global_variables_initializer():

Adds an op to initialize all variables in the model

It seems like both can be used to initialize all variables in graphs. Can we use these two functions exchangbly? If not, what would be the differences?

Answer 1

Unfortunately, you forgot to read an important line in the documentation of tf.initialize_all_variables.

THIS FUNCTION IS DEPRECATED. It will be removed after 2017-03-02. Instructions for updating: Use tf.global_variables_initializer instead.

Answer 2

The changelog for 0.12 version tells you that both both functions do the same thing because:

tf.initialize_all_variables renamed to tf.global_variables_initializer

and as Martianwars mentioned the documentation for initialize_all_variables tells that:

THIS FUNCTION IS DEPRECATED. It will be removed after 2017-03-02. Instructions for updating: Use tf.global_variables_initializer instead.

If you will call it, you will get a warning as well. So you should always use tf.global_variables_initializer()