用TensorFlow写逻辑斯蒂回归

用TensorFlow写逻辑斯蒂回归

具体代码实现

import numpy as np
import tensorflow as tf
import math

# data : num num num num num num num ...num(0||1)
#              ......

class data_reader:
    file_name = str()
    def __init__(self,file_name):
        data_reader.file_name = file_name

    def get_data_list(self,num_lost):
        file_ = open(self.file_name,'r+')
        db = list()       
        for line in file_.readlines():
            reader_list = list()
            for reader_ in line.split():
                try:
                    eval(reader_)
                except:
                    reader_list.append(num_lost)  
                else:
                    reader_list.append(eval(reader_))     
            db.append(reader_list)      
        file_.close()
        return db

    def get_x_y_list(self,db):    
        if type(db) != type(list()):
            return None
        x_list = list()
        y_list = list()
        for reader_list in db:
            length_list = len(reader_list)
            y_list.append(reader_list[length_list-1])
            del reader_list[length_list-1]
            x_list.append(reader_list)
        return x_list,y_list
if __name__ == '__main__':
    # 读取数据
    reader = data_reader('test.dat')
    # 数据初始化
    x_list,y_list = reader.get_x_y_list(reader.get_data_list(num_lost = 1.0))
    # 创建变量
    x_data = tf.constant(np.array(x_list),dtype = tf.float32)
    y_data = tf.constant(np.mat(y_list).transpose(),dtype = tf.float32)
    # 权重的初始化
    Weights = tf.Variable(tf.zeros([len(x_list[0]),1]))
    biase = tf.Variable(tf.zeros([1, 1]))
    # 方程 logistic
    # matmul:矩阵乘法
    # exp : e^x函数
    y = 1.0 / (1 + tf.exp(-tf.matmul(x_data,Weights) - biase))
    # 差值 
    loss = tf.reduce_mean(-(tf.log(y)*y_data + (1 - y_data)*tf.log(1 - y)))
    # 建立优化器
    optimizer = tf.train.GradientDescentOptimizer(0.1)
    # 减少误差，提升准确度
    train = optimizer.minimize(loss)
    # 输入所有的变量
    init = tf.global_variables_initializer()
    # 初始化
    with tf.Session() as sess:
        # 激活神经网络
        sess.run(init)

        # 训练1000步
        for step in range(1001):
            sess.run(train)
            print(step,'\n',sess.run(Weights),sess.run(biase),sess.run(loss))