optimizer.py

import tensorflow as tf
import tensorlayer as tl
import numpy as np

class grader:
    def __init__(self,hidden_size,type,lr,sess,num):
        self.sess = sess
        self.type = type
        with tf.variable_scope(type,reuse=tf.AUTO_REUSE) as scope:
            self.scope = type
            self.cell_list = []
            self.input = tf.placeholder(tf.float32,[num,5,2])
            self.rnn = tl.layers.InputLayer(self.input,name="in")
            self.rnn = tl.layers.RNNLayer(self.rnn,cell_fn = tf.nn.rnn_cell.BasicLSTMCell,n_hidden = hidden_size,n_steps = 1,initializer=tf.random_uniform_initializer(-1, 1),return_last = False,return_seq_2d = False,name="rnn1")
            self.rnn = tl.layers.RNNLayer(self.rnn,cell_fn = tf.nn.rnn_cell.BasicLSTMCell,n_hidden = hidden_size,n_steps = 1,initializer=tf.random_uniform_initializer(-1, 1),return_last = False,return_seq_2d = True,name = "rnn2")
            self.rnn = tl.layers.DenseLayer(self.rnn,n_units=1,W_init=tf.truncated_normal_initializer(stddev=0.1),name = "dense_opti")
            self.output = self.rnn.outputs
            self.output = self.output * 0.1
            #self.W = tf.get_variable("W",[hidden_size,1],dtype=tf.float32)
            #self.b = tf.get_variable("b",[1],dtype = tf.float32)
            #self.tvars =  tf.trainable_variables()
            self.tvars = self.rnn.all_params
            print(self.tvars)
            self.optimizer = tf.train.AdamOptimizer(lr)



    def train(self,loss):
        #with tf.variable_scope(self.type,reuse=tf.AUTO_REUSE) as scope:
        grads = tf.gradients(loss, self.tvars)
        print(grads)
        self.train_op = self.optimizer.apply_gradients(zip(grads, self.tvars))

    def save(self):
        tl.files.exists_or_mkdir(self.scope)
        tl.files.save_ckpt(sess=self.sess, mode_name='model.ckpt', var_list=self.tvars,
                           save_dir=self.scope, printable=False)

    def load(self):
        tl.files.load_ckpt(sess=self.sess, var_list=self.tvars, save_dir=self.scope, printable=False)