Tape not deleted

[asokraju]

In the Critic Class, after accessing the gradients in action_gradient method in lines 150-155

    def action_gradient(self, input_state, input_actions):
        var = tf.constant(input_actions)
        with tf.GradientTape(watch_accessed_variables=False) as tape_a:
            tape_a.watch(var)
            prediction_a = self.critic_model([input_state, var])
        return tape_a.gradient(prediction_a, var)

using GradientTape with watch_accessed_variables=False, the tape_a is not deleted. It may cause undesirable behaviors.

[96RadhikaJadhav]

@asokraju Did you mean to change the code to below?

def action_gradient(self, input_state, input_actions):
var = tf.constant(input_actions)
with tf.GradientTape(watch_accessed_variables=False):
watch(var)
prediction_a = self.critic_model([input_state, var])
return gradient(prediction_a, var)

[asokraju]

I can't seem to notice any changes in the code above. Can you please mark the differences.
I want to just del the tape.

[96RadhikaJadhav]

Changing the above code to below:

[asokraju]

I meant that I forgot to delete the variable tape, after using it. Not delete the word tape in tape_a.

…

On Sat, Jan 16, 2021 at 12:43 PM Radhika ***@***.***> wrote: [image: image] <https://user-images.githubusercontent.com/56536997/104818830-f5892a80-584f-11eb-8e61-47cac84efa8c.png> Changing the above code to below: [image: image] <https://user-images.githubusercontent.com/56536997/104818895-557fd100-5850-11eb-8b38-cb39bc7db5fa.png> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFM53A5BLQ2355TQL32UCELS2HF2VANCNFSM4WBU6UMQ> .

--

________________________________________________ Krishna Chaitanya Kosaraju, PhD Department of Electrical Engineering University of Notre Dame, Indiana, US