深度学习06

循环神经网络

 

 

 

代码

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


#tf.compat.v1.reset_default_graph()
tf.compat.v1.disable_eager_execution()
mnist = input_data.read_data_sets("./mnist_data", one_hot=True)

trainimgs= mnist.train.images
trainlabels = mnist.train.labels
testimgs= mnist.test. images
testlabels = mnist.test.labels
ntrain,ntest,dim, nclasses= trainimgs.shape[0], testimgs.shape[0], trainimgs.shape[1], trainlabels.shape[1]


diminput= 28
dimhidden = 128
dimoutput = nclasses
nsteps= 28
weights = {
    'hidden': tf.Variable(tf.compat.v1.random_normal([diminput, dimhidden])),
    'out': tf.Variable(tf.compat.v1.random_normal([dimhidden,dimoutput]))
}
biases = {
    'hidden': tf.Variable(tf.compat.v1.random_normal([dimhidden])),
    'out': tf.Variable(tf.compat.v1.random_normal ([dimoutput]))
}

def _RNN(_X,_W,_b,_nsteps,_name):
    #[batchsize, nsteps,diminput]
    _X = tf.transpose(_X,[1,0,2])
    #[nsteps*batchsize，diminput]
    _X = tf.reshape(_X, [-1, diminput])
    _H = tf.matmul(_X, _W['hidden']) + _b['hidden']
    _Hsplit = tf.split(_H,_nsteps,0)

    print(_name)
    with tf.compat.v1.variable_scope(_name, reuse=tf.compat.v1.AUTO_REUSE) as scope:
        #变量共享
        scope.reuse_variables()

        lstm_cell= tf.compat.v1.nn.rnn_cell.BasicLSTMCell(dimhidden,forget_bias = 1.0)
        _LSTM_O, _LSTM_S = tf.compat.v1.nn.static_rnn(lstm_cell,_Hsplit, dtype = tf.float32)
    _O = tf.matmul(_LSTM_O[-1],_W['out']) + _b['out']

    return{
    'X': _X,'H': _H, 'Hsplit': _Hsplit,'LSTM_O':_LSTM_O,'LSTM_S':_LSTM_S,'O':_O
    }


learning_rate = 0.001
x= tf.compat.v1.placeholder(tf.float32,[None,nsteps,diminput])
y= tf.compat.v1.placeholder(tf.float32,[None,dimoutput])
myrnn =_RNN(x,weights,biases,nsteps,'basic')
pred= myrnn['O']
cost = tf.reduce_mean(tf.nn. softmax_cross_entropy_with_logits (pred, y))
optm = tf.compat.v1.train.GradientDescentOptimizer(learning_rate).minimize(cost)# Adam
accr = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(pred,1),tf.argmax(y,1)),tf.float32))
init=tf.compat.v1.global_variables_initializer()


with tf.compat.v1.Session() as sess:
    # 初始化变量
    sess.run(init)

    # 开始训练
    for i in range(100):
        # 获取真实值
        batch_xs,batch_ys = mnist.train.next_batch(16)
        batch_xs=batch_xs.reshape((16,nsteps,diminput))
        # print(image.shape)
        # print(label.shape)

        _, loss_value, accuracy_value = sess.run([optm, cost, accr], feed_dict={x: batch_xs, y: batch_ys})
        print("第%d次的损失为%f，准确率为%f" % (i + 1, loss_value, accuracy_value))