import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
# In[3]:
# 载入数据集
mnist = input_data.read_data_sets("MNIST_data", one_hot=True)
# 每个批次的大小
batch_size = 100
# 计算一共有多少个批次
n_batch = mnist.train.num_examples // batch_size
# 定义两个placeholder
x = tf.placeholder(tf.float32, [None, 784])
y = tf.placeholder(tf.float32, [None, 10])
keep_prob =tf.placeholder(tf.float32)
# 创建一个简单的神经网络
W1 = tf.Variable(tf.truncated_normal([784, 2000], stddev=0.1))
b1 = tf.Variable(tf.zeros([2000])+0.1)
L1 = tf.nn.tanh(tf.matmul(x, W1) + b1)
L1_drop = tf.nn.dropout(L1, keep_prob)
W2 = tf.Variable(tf.truncated_normal([2000, 2000], stddev=0.1))
b2= tf.Variable(tf.zeros([2000])+ 0.1)
L2= tf.nn.tanh(tf.matmul(L1_drop, W2) + b2)
L2_drop = tf.nn.dropout(L2, keep_prob)
W3 = tf.Variable(tf.truncated_normal([2000, 1000], stddev=0.1))
b3 = tf.Variable(tf.zeros([1000]) + 0.1)
L3 = tf.nn.tanh(tf.matmul(L2_drop, W3) + b3)
L3_drop = tf.nn.dropout(L3, keep_prob)
W4 = tf.Variable(tf.truncated_normal([1000, 10], stddev=0.1))
b4 = tf.Variable(tf.zeros([10]) + 0.1)
prediction = tf.nn.softmax(tf.matmul(L3_drop, W4) + b2)
# 交叉熵函数
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction))
# 使用梯度下降法
train_step = tf.train.GradientDescentOptimizer(0.2).minimize(loss)
# 初始化变量
init = tf.global_variables_initializer()
# 结果存放在一个布尔型列表中
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1)) # argmax返回一维张量中最大的值所在的位置
# 求准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
with tf.Session() as sess:
sess.run(init)
for epoch in range(31):
for batch in range(n_batch):
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
sess.run(train_step, feed_dict={x: batch_xs, y: batch_ys})
test_acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels, keep_prob: 1.0})
train_acc = sess.run(accuracy, feed_dict={x: mnist.train.images, y: mnist.train.labels, keep_prob: 1.0})
print("Iter " + str(epoch) + ",Testing Accuracy " + str(test_acc)+",Train Accuracy" + str(train_acc))
再迭代300次 3000次都一样的，基本上提高不了了
Iter 0,Testing Accuracy 0.9208,Training Accuracy 0.91416365
Iter 1,Testing Accuracy 0.9298,Training Accuracy 0.9265636
Iter 2,Testing Accuracy 0.9368,Training Accuracy 0.9341818
Iter 3,Testing Accuracy 0.9423,Training Accuracy 0.9419091
Iter 4,Testing Accuracy 0.9439,Training Accuracy 0.94298184
Iter 5,Testing Accuracy 0.9482,Training Accuracy 0.94812727
Iter 6,Testing Accuracy 0.9482,Training Accuracy 0.95054543
Iter 7,Testing Accuracy 0.9521,Training Accuracy 0.95414543
Iter 8,Testing Accuracy 0.9535,Training Accuracy 0.95536363
Iter 9,Testing Accuracy 0.9543,Training Accuracy 0.95714545
Iter 10,Testing Accuracy 0.9569,Training Accuracy 0.9596909
Iter 11,Testing Accuracy 0.956,Training Accuracy 0.96067274
Iter 12,Testing Accuracy 0.9588,Training Accuracy 0.96314543
Iter 13,Testing Accuracy 0.9607,Training Accuracy 0.9644182
Iter 14,Testing Accuracy 0.9607,Training Accuracy 0.9657818
Iter 15,Testing Accuracy 0.9613,Training Accuracy 0.9667818
Iter 16,Testing Accuracy 0.9636,Training Accuracy 0.96834546
Iter 17,Testing Accuracy 0.9636,Training Accuracy 0.9676909
Iter 18,Testing Accuracy 0.9635,Training Accuracy 0.96983635
Iter 19,Testing Accuracy 0.9654,Training Accuracy 0.97056365
Iter 20,Testing Accuracy 0.9667,Training Accuracy 0.97096366
Iter 21,Testing Accuracy 0.9666,Training Accuracy 0.9719091
Iter 22,Testing Accuracy 0.9677,Training Accuracy 0.97285455
Iter 23,Testing Accuracy 0.9666,Training Accuracy 0.9730545
Iter 24,Testing Accuracy 0.968,Training Accuracy 0.9742182
Iter 25,Testing Accuracy 0.9684,Training Accuracy 0.97458184
Iter 26,Testing Accuracy 0.9696,Training Accuracy 0.9758545
Iter 27,Testing Accuracy 0.9688,Training Accuracy 0.9760909
Iter 28,Testing Accuracy 0.9699,Training Accuracy 0.9764364
Iter 29,Testing Accuracy 0.971,Training Accuracy 0.9768909
Iter 30,Testing Accuracy 0.9705,Training Accuracy 0.97776365
使用dropout进行防止过拟合
但是用dropout收敛速度会变慢，到第30次才能达到97.
Iter 0,Testing Accuracy 0.915,Training Accuracy 0.91123635
Iter 1,Testing Accuracy 0.9298,Training Accuracy 0.92745453
Iter 2,Testing Accuracy 0.9366,Training Accuracy 0.9358364
Iter 3,Testing Accuracy 0.9415,Training Accuracy 0.9400909
Iter 4,Testing Accuracy 0.9444,Training Accuracy 0.94485456
Iter 5,Testing Accuracy 0.9461,Training Accuracy 0.9476909
Iter 6,Testing Accuracy 0.9498,Training Accuracy 0.9505091
Iter 7,Testing Accuracy 0.9528,Training Accuracy 0.9532727
Iter 8,Testing Accuracy 0.9531,Training Accuracy 0.9565091
Iter 9,Testing Accuracy 0.955,Training Accuracy 0.9577818
Iter 10,Testing Accuracy 0.9555,Training Accuracy 0.95861816
Iter 11,Testing Accuracy 0.9579,Training Accuracy 0.9612909
Iter 12,Testing Accuracy 0.9595,Training Accuracy 0.96385455
Iter 13,Testing Accuracy 0.9604,Training Accuracy 0.96523637
Iter 14,Testing Accuracy 0.9609,Training Accuracy 0.96592724
Iter 15,Testing Accuracy 0.9611,Training Accuracy 0.96647274
Iter 16,Testing Accuracy 0.9618,Training Accuracy 0.9676727
Iter 17,Testing Accuracy 0.9627,Training Accuracy 0.9693091
Iter 18,Testing Accuracy 0.9642,Training Accuracy 0.96983635
Iter 19,Testing Accuracy 0.9656,Training Accuracy 0.97049093
Iter 20,Testing Accuracy 0.9637,Training Accuracy 0.9705273
Iter 21,Testing Accuracy 0.9659,Training Accuracy 0.97261816
Iter 22,Testing Accuracy 0.9672,Training Accuracy 0.9735818
Iter 23,Testing Accuracy 0.9673,Training Accuracy 0.9735091
Iter 24,Testing Accuracy 0.9677,Training Accuracy 0.97409093
Iter 25,Testing Accuracy 0.9704,Training Accuracy 0.9747818
Iter 26,Testing Accuracy 0.9696,Training Accuracy 0.9760727
Iter 27,Testing Accuracy 0.9699,Training Accuracy 0.9764
Iter 28,Testing Accuracy 0.9687,Training Accuracy 0.9764364
Iter 29,Testing Accuracy 0.968,Training Accuracy 0.9772
Iter 30,Testing Accuracy 0.9707,Training Accuracy 0.97778183
如果使用tensorbored就可以看到曲线图。
为什么要用dropout？
因为从第二段代码我们可以看出，其在测试集的准确率和在训练集的准确率差不多
如果是googlenet和alexnet这么大的网络进行自己分类时，如果我们用goolenet训练500张照片分5类，那么就会过导致过拟合的
因为样本很少，但是我们dropout百分50的神经元那么就会得到好的效果。
用复杂的网络去训练小样本时我们就可以看到dropout的重要性了。