python-tensorflow实现softmax二分类（精度0.71，待改进）

'''
作业：自己构建一个数据，分类的额数据（特性：特征x1和x2,
5*x1-3*x2>0是第一个类别，5*x1-3*x2<0是第二个类别）
使用softmax相关API实现分类
'''
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
import random

'''Tensorflow实现分类任务'''

'''先实现二分类'''
def binary_classification():
    learningRate = 0.005
    #构建分类数据
    x1 = np.arange(-10,10,0.01).reshape(-1,1)
    x2 = np.arange(-10,10,0.01).reshape(-1,1)
    x = np.hstack((x1,x2))
    y  =  3*x1+2*x2**2

    y_label = np.zeros((y.shape[0],2))
    y_label[y[:,0]>100,0] = 1       #0
    y_label[y[:,0]<=100,1] = 1      #1

    #将数据分为2份，一份用于测试，一份用于训练
    random.seed(0)
    data_list = list(range(2000))
    train_list = [random.randint(0,1999) for i in range(int(x1.shape[0]*0.8))]
    #训练数据
    train_x = x[train_list,:]
    train_y = y_label[train_list,:]

    #测试数据
    for i in train_list:
        if i in data_list:
            data_list.remove(i)
    test_list = data_list
    test_x = x[test_list,:]
    test_y = y_label[test_list,:]

    plt.plot(y,y_label,'*')
    # plt.show()

    #构建模型
    xx = tf.placeholder(dtype=tf.float32,shape=[None,2],name='xx')
    yy = tf.placeholder(dtype=tf.float32,shape=[None,2],name='yy')

    layer1_v = tf.Variable(initial_value=tf.random_normal(shape=[2,2],dtype=tf.float32,seed=0),name='layer1')
    b = tf.Variable(initial_value=tf.random_normal(shape=[2],seed=0),name='b')  #加上了bias后精度从0.57提升到了0.71
    
    #softmax进行二分类
    predict = tf.nn.softmax(tf.matmul(xx,layer1_v)+b)
    #损失函数
    loss = tf.reduce_sum(-yy*tf.log(predict))

    #优化器
    Optimizer = tf.train.AdamOptimizer(learning_rate=learningRate).minimize(loss)

    #accuracy
    accuracy = tf.equal(tf.argmax(predict,1),tf.argmax(yy,1))
    accuracy = tf.reduce_mean(tf.cast(accuracy,dtype=tf.float32))

    with tf.Session() as sess:
        tf.global_variables_initializer().run()

        step = 0
        for i in range(1000):
            Optimizer.run(feed_dict={xx:train_x,yy:train_y})
            print("loss:%f,accuracy:%f"%(loss.eval(feed_dict={xx:train_x,yy:train_y}),accuracy.eval(feed_dict={xx:test_x,yy:test_y})))
binary_classification()

精度测试：

0.71 待改进（或许通过批次送入进行训练会增加模型的精度）