vae_imdb.py 文件源码

def build_model(self):

        self.input_y = tf.placeholder(tf.float32, [None,self.num_class], name="input_y") # 1*1, 1doc
        self.one_hot = tf.reshape(tf.cast(tf.one_hot(tf.cast(self.input_y, tf.int32), 2,0,1), tf.float32), [-1,2])


        self.recon_loss = -tf.reduce_sum(tf.log(0.0001 + tf.gather(self.p_xi_h, self.x_id)))
        self.KL = -0.5 * tf.reduce_sum(1.0 + self.hlogvar - tf.pow(self.hmean, 2)\
                  - tf.exp(self.hlogvar), reduction_indices = 1)
        self.loss = tf.reduce_mean(0.0001 * self.KL + self.recon_loss)

        self.optimizer = tf.train.AdamOptimizer(self.learning_rate,0.9)
        self.grads_and_vars = self.optimizer.compute_gradients(self.loss)
        self.capped_gvs = [(tf.clip_by_value(grad, -5., 5.), var) for grad, var in self.grads_and_vars] 
        self.train_op = self.optimizer.apply_gradients(self.capped_gvs)
        #self.optimizer = tf.train.AdamOptimizer(self.learning_rate,beta1=0.9).minimize(self.loss)

        self.init = tf.initialize_all_variables()
        self.sess.run(self.init)