rnn = RNN()
ff = FeedForwardNN()
hidden_state = [0,0,0]
for word in input:
	output,hidden_state = rnn(word,hidden_state)
prediction = ff(output)

tf.nn.rnn_cell.BasicRNNCell(
	num_units, 
	activation=None, 
	reuse=None, 
	name=None, 
	dtype=None, 
	**kwargs)

RNN_cell = tf.nn.rnn_cell.BasicRNNCell(n_hidden_units,activation=tf.nn.tanh)

_init_state =  RNN_cell.zero_state(batch_size,tf.float32)

tf.nn.dynamic_rnn(
    cell,
    inputs,
    sequence_length=None,
    initial_state=None,
    dtype=None,
    parallel_iterations=None,
    swap_memory=False,
    time_major=False,
    scope=None
)

outputs,states = tf.nn.dynamic_rnn(RNN_cell,X_in,initial_state = _init_state,time_major = False)

def RNN(X,weights,biases):
    #X最開始的形狀為(128 batch,28 steps,28 inputs)
    #轉(zhuǎn)化為(128 batch*28 steps,128 hidden)
    X = tf.reshape(X,[-1,n_inputs])
    #經(jīng)過乘法后結(jié)果為(128 batch*28 steps,256 hidden)
    X_in = tf.matmul(X,weights['in'])+biases['in']
    #再次轉(zhuǎn)化為(128 batch,28 steps,256 hidden) 
    X_in = tf.reshape(X_in,[-1,n_steps,n_hidden_units])
    RNN_cell = tf.nn.rnn_cell.BasicRNNCell(n_hidden_units,activation=tf.nn.tanh)
    _init_state =  RNN_cell.zero_state(batch_size,tf.float32)
    outputs,states = tf.nn.dynamic_rnn(RNN_cell,X_in,initial_state = _init_state,time_major = False)
    results = tf.matmul(states,weights['out'])+biases['out']
    return results

import tensorflow as tf 
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data",one_hot = "true")
lr = 0.001          #學習率
training_iters = 1000000        #學習世代數(shù)
batch_size = 128                #每一輪進入訓練的訓練量
n_inputs = 28                   #輸入每一個隱含層的inputs維度
n_steps = 28                    #一共分為28次輸入
n_hidden_units = 128           #每一個隱含層的神經(jīng)元個數(shù)
n_classes = 10                  #輸出共有10個
x = tf.placeholder(tf.float32,[None,n_steps,n_inputs])
y = tf.placeholder(tf.float32,[None,n_classes])
weights = {
    'in':tf.Variable(tf.random_normal([n_inputs,n_hidden_units])),
    'out':tf.Variable(tf.random_normal([n_hidden_units,n_classes]))
}
biases = {
    'in':tf.Variable(tf.constant(0.1,shape=[n_hidden_units])),
    'out':tf.Variable(tf.constant(0.1,shape=[n_classes]))
}
def RNN(X,weights,biases):
    #X最開始的形狀為(128 batch,28 steps,28 inputs)
    #轉(zhuǎn)化為(128 batch*28 steps,128 hidden)
    X = tf.reshape(X,[-1,n_inputs])
    #經(jīng)過乘法后結(jié)果為(128 batch*28 steps,256 hidden)
    X_in = tf.matmul(X,weights['in'])+biases['in']
    #再次轉(zhuǎn)化為(128 batch,28 steps,256 hidden) 
    X_in = tf.reshape(X_in,[-1,n_steps,n_hidden_units])
    RNN_cell = tf.nn.rnn_cell.BasicRNNCell(n_hidden_units,activation=tf.nn.tanh)
    _init_state =  RNN_cell.zero_state(batch_size,tf.float32)
    outputs,states = tf.nn.dynamic_rnn(RNN_cell,X_in,initial_state = _init_state,time_major = False)
    results = tf.matmul(states,weights['out'])+biases['out']
    return results
pre = RNN(x,weights,biases)
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = pre,labels = y))
train_op = tf.train.AdamOptimizer(lr).minimize(cost)
correct_pre = tf.equal(tf.argmax(y,1),tf.argmax(pre,1))
accuracy = tf.reduce_mean(tf.cast(correct_pre,tf.float32))
init = tf.initialize_all_variables()
with tf.Session() as sess:
    sess.run(init)
    step = 0
    while step*batch_size <training_iters:
        batch_xs,batch_ys = mnist.train.next_batch(batch_size)
        batch_xs = batch_xs.reshape([batch_size,n_steps,n_inputs])
        sess.run(train_op,feed_dict = {
            x:batch_xs,
            y:batch_ys
        })
        if step%20 == 0:
            print(sess.run(accuracy,feed_dict = {
                x:batch_xs,
                y:batch_ys
            }))
        step += 1