————————————————————————————————————
写在开头:此文参照莫烦python教程(墙裂推荐!!!)
————————————————————————————————————
循环神经网络RNN
相关名词:
- LSTM:长短期记忆
- 梯度消失/梯度离散
- 梯度爆炸
- 输入控制:控制是否把当前记忆加入主线网络
- 忘记控制:控制是否暂时忘记主线网络,先看当前分线
- 输出控制: 控制输出是否要考虑要素
- 数据有顺序的/序列化
- 前面的影响后面的
RNN LSTM 之分类
识别手写数字
- 识别手写数字
- mnist数据集
- 一行一行地识别
rnn使用错误及修正
- 错误一:
错误描述: ValueError: Variable tf.nn.dynsmic_rnn/rnn/basic_lstm_cell/kernel already exists, disallowed. Did you mean to set reuse=True or reuse=tf.AUTO_REUSE in VarScope? Originally defined at:
错误解决:看你的训练数据和测试数据是否放在同一个文件下,若是,要加上下面一句:
#如果训练和测试数据存放在同一个文件中,一定要加下面这句!
tf.reset_default_graph()
如果这时候出现了错误二,就用下面的解决方法:
- 错误二
错误描述:ValueError: Tensor(“tf.nn.dynsmic_rnn/rnn/Const:0”, shape=(1,), dtype=int32) must be from the same graph as Tensor(“ExpandDims:0”, shape=(1,), dtype=int32).
错误解决:
#把tf.reset_default_graph() 改为:
tf.Graph()
完整代码
下面是完整的分类代码及结果
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
#load data
mnist = input_data.read_data_sets('MNIST_data',one_hot = True)
#参数
lr = 0.001
training_iters = 100000 #循环次数
batch_size = 128
n_inputs = 28 #因为照片是28*28,而每次都读一行,所以input为28
n_steps = 28 #因为有28行,所以要input28步
n_hidden_unis = 128 #隐藏层,自己设
n_classes = 10 #10个数字(0-9),所以类别有10种
#holder
x = tf.placeholder(tf.float32,[None,n_steps,n_inputs])
y = tf.placeholder(tf.float32,[None,n_classes])
#定义权重
weights = {
#input weights(28,128)
'in':tf.Variable(tf.random_normal([n_inputs,n_hidden_unis])),
#output weights(128,10)
'out':tf.Variable(tf.random_normal([n_hidden_unis,n_classes]))
}
#定义偏置
biases = {
'in':tf.Variable(tf.constant(0.1,shape=[n_hidden_unis,])),
'out':tf.Variable(tf.constant(0.1,shape =[n_classes,]))
}
#定义RNN
def RNN(X,weights,biasis):
#hidden layer
#X(128,28,28) ==>(128*28,28)
X = tf.reshape(X,[-1,n_inputs])
X_in =tf.matmul(X,weights['in']+biases['in']) #(128*28,128)
X_in = tf.reshape(X_in,[-1,n_steps,n_hidden_unis])#(128,28,128)
#cell
#forget_bais推荐初始化为1.0
#with tf.variable_scope('lstm_cell'):
lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(n_hidden_unis,forget_bias=1.0,state_is_tuple=True)
#with tf.variable_scope('init_'):
init_state = lstm_cell.zero_state(batch_size,dtype=tf.float32)
#output是个列表;这里实践维度在行,就是X_in的第二个,所以为false,时间维度为第一个,则true
with tf.variable_scope('tf.nn.dynsmic_rnn'):
outputs,states = tf.nn.dynamic_rnn(lstm_cell,X_in,initial_state=init_state,time_major=False)
#output
results = tf.matmul(states[1],weights['out']+biases['out'])
##other way,这里可用
#outputs = tf.unpack(tf.transpose(outputs,[1,0,2]))
#results = tf.matmuo(outputs[-1],weights['out']+biases['out'])
return results
#如果训练和测试数据存放在同一个文件中,一定要加下面这句!
tf.Graph()
pred = RNN(x,weights,biases)
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=pred))
train_op = tf.train.AdamOptimizer(lr).minimize(cost)
correct_pred = tf.equal(tf.argmax(pred,1),tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred,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%50 == 0:
print(sess.run(accuracy,feed_dict = {
x:batch_xs,y:batch_ys,
}))
step += 1
Extracting MNIST_data\train-images-idx3-ubyte.gz
Extracting MNIST_data\train-labels-idx1-ubyte.gz
Extracting MNIST_data\t10k-images-idx3-ubyte.gz
Extracting MNIST_data\t10k-labels-idx1-ubyte.gz
0.2109375
0.78125
0.84375
0.9140625
0.921875
0.921875
0.9375
0.9453125
0.96875
0.9140625
0.953125
0.984375
0.9609375
0.9453125
0.96875
0.9921875
由上面的结果来看,RNN的效果还是很不错的!
*点击[这儿:TensorFlow]发现更多关于TensorFlow的文章*