摘自《TensorflowGoogle实战》
一、网络示意
二、代码
1、创建网络
1 import tensorflow as tf 2 from numpy.random import RandomState 3 from sklearn.model_selection import train_test_split 4 5 batch_size = 8 6 7 # 定义网络 8 w1 = tf.Variable(tf.random_normal([2, 3], stddev=1, seed=1)) 9 b1 = tf.Variable(tf.zeros([1, 3], name="bias1")) 10 b2 = tf.Variable(tf.zeros([1], name="bias2")) 11 w2 = tf.Variable(tf.random_normal([3, 1], stddev=1, seed=1)) 12 13 x = tf.placeholder(tf.float32, shape=(None, 2), name='x-input') 14 y_ = tf.placeholder(tf.float32, shape=(None, 1), name='y-input') 15 a = tf.matmul(x, w1) + b1 16 17 # 这里一定不要写成 tf.sigmoid 18 y_last = tf.nn.sigmoid(tf.matmul(a, w2) + b2)
2、定义损失函数
1 # 损失与准确率 2 loss = tf.losses.sigmoid_cross_entropy(y_, y_last) 3 train_step = tf.train.AdamOptimizer(0.07).minimize(loss) 4 5 correct_prediction = tf.equal(tf.round(y_last), y_) 6 acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
3、造数据
1 # 造数据，X为二维数据例如：[[0.1, 0.8]]， 当X的第一项和第二项相加 < 1 时 Y为1， 当X的第一项和第二项相加 >= 1时为 0 2 rdm = RandomState(1) 3 data_set_size = 128000 4 X = rdm.rand(data_set_size, 2) 5 Y = [[int(x1 + x2 < 1)] for (x1, x2) in X] 6 7 X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.3)
4、训练与测试
1 with tf.Session() as sess: 2 init_op = tf.global_variables_initializer() 3 sess.run(init_op) 4 STEPS = 7000 5 for i in range(0, 8900 - batch_size, batch_size): 6 start = i 7 sess.run(train_step, feed_dict={x: X_train[start: start + batch_size], y_: y_train[start: start + batch_size]}) 8 9 if i % 800 == 0: 10 # 计算所有数据的交叉熵 11 total_cross_entropy = sess.run(loss, feed_dict={x: X, y_: Y}) 12 # 输出交叉熵之和 13 # print("After %d training step(s),cross entropy on all data is %g" % (i, total_cross_entropy)) 14 15 acc_ = sess.run(acc, feed_dict={x: X_test, y_: y_test}) 16 print("accuracy on test data is ", acc_) 17 18 test = sess.run(y_last, feed_dict={x: X_test, y_: y_test})
5、结果（个人认为tf有很多bug，有几次结果居然是0.5左右，不知道为啥）
accuracy on test data is 0.96091145