I am trying to create a diagonal matrix using tf.get_variable But I do not know how! Like I can make a variable which is a diagonal matrix like:
dia_size = tf.zeros((num_filters, img_size))
b = tf.Variable(tf.matrix_diag(dia_size), name=name)
b = tf.reshape(b, [-1, img_size, img_size, num_filters])
but I can not do it with tf.get_variable.
Thanks for your help in advance!
If you set the initializer
parameter of tf.get_variable
to a tensor, the variable will be initialized to the tensor's value. Therefore, you can use the following code:
dia_size = tf.zeros((num_filters, img_size))
b = tf.matrix_diag(dia_size)
var = tf.get_variable(..., initializer=b, ...)