0a1844fe38592ba9797e37c920d6a9f745f217c0,async_knowledge_transfer.py,AsyncKnowledgeTransferLearner,build_networks,#AsyncKnowledgeTransferLearner#,197

Before Change

        self.action_taken = tf.placeholder(tf.float32, name="action_taken")
        self.advantage = tf.placeholder(tf.float32, name="advantage")

        W0 = tf.Variable(tf.random_normal([self.nO, self.config["n_hidden_units"]]) / np.sqrt(self.nO), name="W0")
        b0 = tf.Variable(tf.zeros([self.config["n_hidden_units"]]), name="b0")
        self.L1 = tf.tanh(tf.nn.xw_plus_b(self.states, W0, b0), name="L1")

        self.knowledge_base = tf.Variable(tf.random_normal([self.config["n_hidden_units"], self.config["n_sparse_units"]]))

        self.shared_vars = [W0, b0, self.knowledge_base]

        self.optimizer = tf.train.RMSPropOptimizer(learning_rate=self.config["learning_rate"], decay=self.config["decay"], epsilon=self.config["epsilon"])

    def signal_handler(self, signal, frame):

After Change

    def build_networks(self):
        with tf.variable_scope("shared"):
            self.states = tf.placeholder(tf.float32, [None, self.nO], name="states")
            self.action_taken = tf.placeholder(tf.float32, name="action_taken")
            self.advantage = tf.placeholder(tf.float32, name="advantage")

            self.L1 = tf.contrib.layers.fully_connected(
                inputs=self.states,
                num_outputs=self.config["n_hidden_units"],
                activation_fn=tf.tanh,
                weights_initializer=tf.random_normal_initializer(),
                biases_initializer=tf.zeros_initializer(),
                scope="L1")

            self.knowledge_base = tf.Variable(tf.random_normal([self.config["n_hidden_units"], self.config["n_sparse_units"]]), name="knowledge_base")

            self.shared_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope="shared")

Instances