本文使用的tensorflow版本:1.4
tensorflow安装:pip install tensorflow
1、什么是drop-out
dropout是指在深度学习网络的训练过程中,对于神经网络单元,按照一定的概率将其暂时从网络中丢弃。注意是暂时,对于随机梯度下降来说,由于是随机丢弃,故而每一个mini-batch都在训练不同的网络。
dropout是CNN中防止过拟合提高效果的一个大杀器,但对于其为何有效,却众说纷纭。
具体的细节可以参考博客:http://blog.csdn.net/stdcoutzyx/article/details/49022443,这篇写的挺好的,大家可以深入研究一下
2、tensorflow中drop-out的实现
tensorflow中的drop-out非常容易实现,使用下面的语法:
Wx_plus_b = tf.nn.dropout(Wx_plus_b, keep_prob)
此时我们需要定义一个keep_prob的placeholder
keep_prob = tf.placeholder(tf.float32)
当然不要忘记在feed_dict里面加入我们的keep_prob
sess.run(train_step,feed_dict={xs:trainx,ys:trainy,keep_prob:0.5})
3、一个完整的例子
import tensorflow as tf
from sklearn.preprocessing import LabelBinarizer
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_digits
digits = load_digits()
X = digits.data
Y = digits.target
Y = LabelBinarizer().fit_transform(Y)
print(Y.shape)
trainx,testx,trainy,testy = train_test_split(X,Y,test_size=0.3)
def add_layer(input,input_size,output_size,layer_name,activation_function=None):
Weights = tf.Variable(tf.random_normal([input_size,output_size]))
biases = tf.Variable(tf.zeros([1,output_size])+0,1)
Wx_plus_bias = tf.add(tf.matmul(input,Weights),biases)
Wx_plus_bias = tf.nn.dropout(Wx_plus_bias,keep_prob)
if activation_function == None:
outputs = Wx_plus_bias
else:
outputs = activation_function(Wx_plus_bias)
#这里的output是一个二维的,所以每一步对应一个线(或者说小的矩形,颜色越深的地方表示这个地方的数越多,可以认为纵向上表示train到这一步的时候的一个数据分布
tf.summary.histogram(layer_name+'/outputs',outputs)
return outputs
xs = tf.placeholder(tf.float32,[None,64])
ys = tf.placeholder(tf.float32,[None,10])
keep_prob = tf.placeholder(tf.float32)
l1 = add_layer(xs,64,50,"l1",activation_function=tf.nn.tanh)
prediction = add_layer(l1,50,10,"l2",activation_function=tf.nn.softmax)
cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys * tf.log(prediction),reduction_indices=[1]))
#因为cross_entropy是一个标量,所以定义tf.summary.scalar
tf.summary.scalar("loss",cross_entropy)
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
init = tf.global_variables_initializer()
with tf.Session() as sess:
#合并所有的summary
merged = tf.summary.merge_all()
#得到summary的FileWriter
train_writer = tf.summary.FileWriter('logs/train/',sess.graph)
test_writer = tf.summary.FileWriter('logs/test/',sess.graph)
sess.run(init)
for i in range(1000):
sess.run(train_step,feed_dict={xs:trainx,ys:trainy,keep_prob:0.5})
if i % 50 == 0:
#print(sess.run(cross_entropy,feed_dict={xs:trainx,ys:trainy}))
#这里要运行merged
train_loss = sess.run(merged,feed_dict = {xs:trainx,ys:trainy,keep_prob:0.5})
test_loss = sess.run(merged,feed_dict={xs:testx,ys:testy,keep_prob:0.5})
#将loss写入FileWriter中
train_writer.add_summary(train_loss,i)
test_writer.add_summary(test_loss,i)