一个有激活函数和没有激活函数的例子

最新推荐文章于 2025-01-29 14:02:26 发布

原创最新推荐文章于 2025-01-29 14:02:26 发布 · 1.1k 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#tensorflow

神经网络学习专栏收录该内容

49 篇文章

订阅专栏

本文介绍了一种使用TensorFlow实现线性和非线性分类的方法，通过生成和可视化两类不同的测试数据，包括简单的线性分类数据和环形分布数据，展示了如何构建和训练两种模型：一种用于线性分类，另一种用于处理更复杂的非线性分类任务。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

闲来无事写的，本来想可视化一些，但是python编程实在不会，只能简化一下，自己理解吧
PS:环形测试数据的生成代码来自网上

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
#为了说明问题，两个模型都设计为两层隐藏层
tf.reset_default_graph()
#生成测试数据类型1
def create_data1(N):#生成简单线性分类测试数据
    tmpl=np.zeros([N,2])
    mean=np.random.randn(2)
    t1=np.random.multivariate_normal(mean=mean,cov=[[1,0],[0,1]],size=int(N/2))
    t2 = np.random.multivariate_normal(mean=mean+3, cov=[[1, 0], [0, 1]], size=int(N/2))
    tmp1=np.concatenate([t1,t2])
    tmp2=np.zeros(N)
    tmp2[int(N/2):]=1
    tmp2=np.reshape(tmp2,[-1,1])
    return tmp1,tmp2
def create_data2(N):#生成环形测试数据
    tmp1=np.zeros([N,2])
    tmp2=np.zeros(N)
    N = int(N / 2)
    PI=3.1415
    theta = np.linspace(0, 2 * PI, N) + PI * (np.random.rand(N)) / 100
    a = 0.5
    b = 0.5
    r1 = 0.4 + 2 * (np.random.rand(N) - 0.5) / 10
    x1 = a + r1 * np.cos(theta) + (np.random.rand(N) - 0.5) / 50
    y1 = b + r1 * np.sin(theta) + (np.random.rand(N) - 0.5) / 50
    r2 = 0.2 * np.random.rand(N)
    x2 = a + r2 * np.cos(theta) + (np.random.rand(N) - 0.5) / 50
    y2 = b + r2 * np.sin(theta) + (np.random.rand(N) - 0.5) / 50
    tmp1[0:N,0]=x1[:]
    tmp1[0:N, 1] = y1[:]
    tmp1[N:,0]=x2[:]
    tmp1[N:, 1] = y2[:]
    tmp2[N:] = 1
    tmp2 = np.reshape(tmp2, [-1, 1])
    return tmp1,tmp2
data_rows=1000
train_data,train_label=create_data1(data_rows)#这个测试数据可以是data2

plt.subplot(2,1,1)
train_data1,train_label1=create_data1(data_rows)
plt.plot(train_data1[0:500,0],train_data1[0:500,1],'ro')
plt.plot(train_data1[500:,0],train_data1[500:,1],'bo')
plt.subplot(2,1,2)
train_data2,train_label2=create_data2(data_rows)
plt.plot(train_data2[0:500,0],train_data2[0:500,1],'ro')
plt.plot(train_data2[500:,0],train_data2[500:,1],'bo')
plt.show()
#线性的模型
g_line_function=tf.Graph()
with g_line_function.as_default():
    X=tf.placeholder(tf.float32,[None,2])
    Z=tf.placeholder(tf.float32,[None,1])

    Wih=tf.Variable(tf.random_normal([2,2]))
    bih=tf.Variable(tf.zeros([2]))

    Who=tf.Variable(tf.random_normal([2,1]))
    bho=tf.Variable(tf.zeros([1]))

    Zih=tf.matmul(X,Wih)+bih
    Zho=tf.matmul(Zih,Who)+bho

    cost=tf.reduce_mean(tf.square(Z-Zho))
    learning_rate=0.001
    optimizer=tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)

g_active_function=tf.Graph()
with g_active_function.as_default():
    X=tf.placeholder(tf.float32,[None,2])
    Z=tf.placeholder(tf.float32,[None,1])

    Wih=tf.Variable(tf.random_normal([2,2]))
    bih=tf.Variable(tf.zeros([2]))

    Who=tf.Variable(tf.random_normal([2,1]))
    bho=tf.Variable(tf.zeros([1]))

    Zih=tf.nn.sigmoid(tf.matmul(X,Wih)+bih)
    Zho=tf.nn.sigmoid(tf.matmul(Zih,Who)+bho)

    cost=tf.reduce_mean(tf.square(Z-Zho))
    learning_rate=0.001
    optimizer=tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)
with tf.Session(graph=g_active_function) as sess:
    sess.run(tf.global_variables_initializer())
    epochs=200
    batch_size=10
    batchs = int(data_rows/batch_size)
    for epoch in range(epochs):
        avg_loss=0
        for i in range(batchs):
            x=train_data[i*batch_size:(i+1)*batch_size,:]
            z=train_label[i*batch_size:(i+1)*batch_size]
            _,c=sess.run([optimizer,cost],feed_dict={X:x,Z:z})
            zho=sess.run(Zho,feed_dict={X:x,Z:z})
            avg_loss+=c/batchs
        print("Epoch=",epoch,"Loss=",avg_loss)
    print("Finished")