学了一段时间tensorflow以后,对于动手写代码的能力还是不行,看了下莫烦的视频,记录一下。
是一个利用rnn来处理mnist数据集的例子。
分为三部分:第一部分定义参数,第二部分定义网络,第三部分训练
首先定义参数
import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets('MNIST_data',one_hot=True) HIDDEN_LAYER = 128 BATCH_SIZE = 128 INPUTS_NUM = 28 STEPS_NUM = 28 CLASS_NUM = 10 lr = 0.001 traning_iter = 100000
注意lr学习率最开始设置为0.1 收敛速度特别慢。还要注意的就是x,y weights biases的维度问题。x = tf.placeholder(tf.float32,shape=[None,STEPS_NUM,INPUTS_NUM]) y = tf.placeholder(tf.float32,shape=[None,CLASS_NUM]) weights = { 'in':tf.Variable(tf.random_normal([INPUTS_NUM,HIDDEN_LAYER])), 'out':tf.Variable(tf.random_normal(([HIDDEN_LAYER,CLASS_NUM]))) } biases = { 'in':tf.Variable(tf.random_normal([HIDDEN_LAYER,])), 'out':tf.Variable(tf.random_normal(([CLASS_NUM,]))) }
第二步定义的就是神经网络模型,最终是要返回一个结果。
def Rnn(X,weights,biases): X = tf.reshape(X,[BATCH_SIZE*STEPS_NUM,INPUTS_NUM])#? X_in = tf.matmul(X,weights['in'])+biases['in']
X_in = tf.reshape(X_in,[-1,STEPS_NUM,HIDDEN_LAYER])#?
##以上为输入层,最开始的X是【128,28,28】,我们需要先转换为[128*28,28]然后和weights【’inx‘】相乘,然后结果再reshape为HiDDenLayer
cell = tf.nn.rnn_cell.BasicLSTMCell(HIDDEN_LAYER,forget_bias=1.0,state_is_tuple=True) _init_state = cell.zero_state(BATCH_SIZE,tf.float32) output,states = tf.nn.dynamic_rnn(cell,X_in,initial_state=_init_state,time_major=False) result = tf.matmul(states[1],weights['out'])+biases['out'] # result = None return result
最后将结果返回 再训练
pred = Rnn(x,weights,biases) cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred,labels= y)) train_op = tf.train.AdamOptimizer(lr).minimize(cost) correct = tf.equal(tf.argmax(pred,1),tf.argmax(y,1)) accuracy = tf.reduce_mean(tf.cast(correct,dtype=tf.float32)) init = tf.initialize_all_variables() with tf.Session() as sess: sess.run(init) step = 0 while step*BATCH_SIZE<traning_iter: batch_xs,batch_ys = mnist.train.next_batch(BATCH_SIZE) # batch_xs = tf.reshape(batch_xs,[BATCH_SIZE,STEPS_NUM,INPUTS_NUM])#?? batch_xs = batch_xs.reshape([BATCH_SIZE,STEPS_NUM,INPUTS_NUM]) 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 = step+1
产生的问题就是batch_xs最开始是一长串,在将其reshape的过程中,用的tf.shape
完整代码如下
import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets('MNIST_data',one_hot=True) HIDDEN_LAYER = 128 BATCH_SIZE = 128 INPUTS_NUM = 28 STEPS_NUM = 28 CLASS_NUM = 10 lr = 0.001 traning_iter = 100000 x = tf.placeholder(tf.float32,shape=[None,INPUTS_NUM,STEPS_NUM]) y = tf.placeholder(tf.float32,shape=[None,CLASS_NUM]) weights = { 'in':tf.Variable(tf.random_normal([INPUTS_NUM,HIDDEN_LAYER])), 'out':tf.Variable(tf.random_normal(([HIDDEN_LAYER,CLASS_NUM]))) } biases = { 'in':tf.Variable(tf.random_normal([HIDDEN_LAYER,])), 'out':tf.Variable(tf.random_normal(([CLASS_NUM,]))) } def Rnn(X,weights,biases): X = tf.reshape(X,[BATCH_SIZE*STEPS_NUM,INPUTS_NUM])#? X_in = tf.matmul(X,weights['in'])+biases['in'] X_in = tf.reshape(X_in,[-1,STEPS_NUM,HIDDEN_LAYER])#? cell = tf.nn.rnn_cell.BasicLSTMCell(HIDDEN_LAYER,forget_bias=1.0,state_is_tuple=True) _init_state = cell.zero_state(BATCH_SIZE,tf.float32) output,states = tf.nn.dynamic_rnn(cell,X_in,initial_state=_init_state,time_major=False) result = tf.matmul(states[1],weights['out'])+biases['out'] # result = None return result pred = Rnn(x,weights,biases) cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred,labels= y)) train_op = tf.train.AdamOptimizer(lr).minimize(cost) correct = tf.equal(tf.argmax(pred,1),tf.argmax(y,1)) accuracy = tf.reduce_mean(tf.cast(correct,dtype=tf.float32)) init = tf.initialize_all_variables() with tf.Session() as sess: sess.run(init) step = 0 while step*BATCH_SIZE<traning_iter: batch_xs,batch_ys = mnist.train.next_batch(BATCH_SIZE) # batch_xs = tf.reshape(batch_xs,[BATCH_SIZE,STEPS_NUM,INPUTS_NUM])#?? batch_xs = batch_xs.reshape([BATCH_SIZE,STEPS_NUM,INPUTS_NUM]) 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 = step+1