Tensorflow 实战：一个简单的二分类模型

自己动手建一个二分类的网络模型，所有代码都是自己一个字母一个字母敲进去的。训练数据正样本是半径为 4 、圆心在原点的园内。我建了一个两层的神经网络，两个输入节点(x0, x1)代表平面坐标，一个输出 y 代表正负样本标识。代码如下：

import tensorflow as tf
import numpy as np

TRAIN_SIZE = 100000
TEST_SIZE = 1000
LARYER1_SIZE = 16
CIRCLE_R = 4.0

training_epochs = 10000
learning_rate = 2.0

#生成训练数据
train_x = (np.random.rand(TRAIN_SIZE, 2) - 0.5) * 10
train_y = np.empty([TRAIN_SIZE, 1], dtype = float) 
for i in range(TRAIN_SIZE):
    x0 = train_x[i, 0]
    x1 = train_x[i, 1]
    if x0*x0 + x1*x1 < CIRCLE_R * CIRCLE_R:
        train_y[i] = 1.0
    else:
        train_y[i] = 0.0

#生成测试数据
test_x = (np.random.rand(TEST_SIZE, 2) - 0.5) * 10
test_y = np.empty([TEST_SIZE, 1], dtype = float) 
for i in range(TEST_SIZE):
    x0 = test_x[i, 0]
    x1 = test_x[i, 1]
    if x0*x0 + x1*x1 < CIRCLE_R * CIRCLE_R:
        test_y[i] = 1.0
    else:
        test_y[i] = 0.0

#定义模型
x = tf.placeholder(tf.float32, [None, 2])
y = tf.placeholder(tf.float32, [None, 1])
w1 = tf.Variable(tf.random_normal([2, LARYER1_SIZE]))
b1 = tf.Variable(tf.zeros([1, LARYER1_SIZE]))
w2 = tf.Variable(tf.random_normal([LARYER1_SIZE, 1]))
b2 = tf.Variable(tf.zeros([1]))
layer1 = tf.sigmoid(tf.matmul(x, w1) + b1)
pred = tf.sigmoid(tf.matmul(layer1, w2) + b2)

cost = tf.reduce_mean(tf.abs(y-pred))
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for i in range(training_epochs):
        _, c = sess.run([optimizer, cost], feed_dict = {x: train_x, y: train_y})
        print ("Epoch:", "%04d" % (i+1), "cost = ", "{:9f}".format(c))
    print("finished!")

    correct_prediction = tf.equal(tf.greater(pred, 0.5), tf.greater(y, 0.5))  
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    print("Accuracy_train: ", accuracy.eval({x: train_x, y: train_y}))
    print("Accuracy_test : ", accuracy.eval({x: test_x, y: test_y}))   

运行结果如下：

... ...
Epoch: 9997 cost =   0.023868
Epoch: 9998 cost =   0.023867
Epoch: 9999 cost =   0.023866
Epoch: 10000 cost =   0.023865
finished!
Accuracy_train:  0.99499995
Accuracy_test :  0.99600005

第一次实战，趟了很多坑。我觉得自己用 C 语言写代码，都会比用 Tensorflow 效率高。不过作为最流行的深度学习框架，还是必须熟练掌握的。这几年时间都浪费在非代码工作上了，心疼呀！