Get_variable with initializer xavier

tensorflow

#1

I’m trying to emulate a python call to get_variable() in Julia:
Python:
tf.get_variable(name,size,initializer=tf.contrib.layers.xavier_initializer(seed = 0))
Julia:
get_variable(size,…?..)

but not getting the xavier initializer to work. I see that Knet offers an xavier() function but does not seem to fit in with this statement.

A search in the docs http://malmaud.github.io/TensorFlow.jl/latest/search.html?q=xavier pulls up two possibilities but no xavier mentioned. What am I missing?


#2

OK so a bit closer with variable_scope() initialization, but no closer to xavier unless I need to provide a custom function.