be3f3039c92721bb54de1a7eda210a87f19da67d,snake/solver/dqn.py,DQNSolver,__build_net,#DQNSolver#,45

Before Change

        self.__q_nxt_all = __build_layers(self.__state_nxt, "target_net", w_init, b_init)

        with tf.variable_scope("q_target"):
            max_actions = tf.argmax(self.__q_eval_all, axis=1, output_type=tf.int32, name="max_actions")
            action_indices = tf.stack([indices, max_actions], axis=1)
            // Shape: (None, )
            q_target = self.__reward + self.__gamma * \
                       tf.gather_nd(self.__q_nxt_all, action_indices, name="gather_nd")
            self.__q_target = tf.stop_gradient(q_target)

        with tf.variable_scope("loss"):
            self.__loss = tf.reduce_mean(tf.squared_difference(

After Change

        self.__q_target = tf.placeholder(
            tf.float32, [None, ], name="q_target")

        scope_eval, scope_target = "eval_net", "target_net"
        w_init = tf.random_normal_initializer(mean=0, stddev=0.3)
        b_init = tf.constant_initializer(0.1)

        // Eval net output / Shape: (None, num_actions)

Instances