NoisyDense incorrect sigma init?
pedrohpf opened this issue · comments
You have this code when init sigma:
sigma_initializer = tf.constant_initializer(self.std_init / np.sqrt(self.units))
Going into the original paper (section 3.2) I would assume the init to be like this:
sigma_initializer = tf.constant_initializer(self.std_init / np.sqrt(input_dim))
Was this change intended? Or maybe I'm misunderstanding the paper?