447885e15243dd18d906e2e35ac34ec6dcf9a600,Reinforcement_learning_TUT/7_Policy_gradient/RL_brain.py,PolicyGradient,_build_net,#PolicyGradient#,129

Before Change

        with tf.variable_scope("eval_net"):
            self.q_eval = self._build_layers(self.s, self.n_actions, trainable=True)
            with tf.name_scope("loss"):
                self.loss = tf.reduce_sum(tf.square(self.q_target - self.q_eval))
            with tf.name_scope("train"):
                self._train_op = tf.train.RMSPropOptimizer(self.lr).minimize(self.loss)

After Change

        l1 = self._add_layer("hidden0", self.x_inputs, self.n_features, 10, tf.nn.relu)     // hidden layer 1
        self.prediction = self._add_layer("output", l1, 10, 1, tf.nn.sigmoid)  // predicting for action 0
        with tf.name_scope("loss"):
            loglik = self.fake_targets*tf.log(self.prediction) + (1 - self.fake_targets)*tf.log(1-self.prediction)  //
            self.loss = -tf.reduce_mean(loglik * self.advantages)
        with tf.name_scope("train"):
            self._train_op = tf.train.RMSPropOptimizer(self.lr).minimize(self.loss)

Instances