a1926719a8174e17f9b002c6369880f11b5dab88,tensorforce/value_functions/deep_q_network.py,DeepQNetwork,create_training_operations,#DeepQNetwork#,121

Before Change

                                          name="target_values")

        with tf.name_scope("training"):
            action_indices = tf.squeeze(np.dstack([tf.range(self.batch_size), self.actions]))
            batch_q_values = tf.identity(self.training_network, name="batch_q_values")
            q_values_actions_taken = tf.gather_nd(batch_q_values, action_indices)

After Change

        with tf.name_scope("training"):
            self.q_targets = tf.placeholder(tf.float32, [None], name="q_targets")
            self.q_values_actions_taken = tf.placeholder(tf.float32, [None], name="q_values_actions_taken")

            // Mean squared error
            loss = tf.reduce_mean(tf.square(self.q_targets - self.q_values_actions_taken), name="loss")
            grads_and_vars = self.optimizer.compute_gradients(loss)

Instances