Tensorflow多元一次回归示例

Tensorflow多元一次回归示例

下面的示例对：y = w1.x1 + w2.x2 + … +wn.xk + b 多元一次方程进行线性回归

#源码

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt


def normalize(X):
	"""
	归一化函数
    :param X:
    :return: x
    """
    mean = np.mean(X)
	std = np.std(X)
	X = (X - mean)/std
	return X


def appendBiasReshape(features_data, k, labels):
    """
    对训练特征进行整形：np.reshape(np.c_[np.ones(m),features],[m,k+1])  来源于对将多元一次方程式的表达：
    w1.x1 + w2.x2 + ... +wn.xk + b = y  => (1,x1,x2,x3,...,xk)(b,w1,w2,w3,...,wk)' = y
    => (x0,x1,x2,x3,...,xk)((w0,w1,w2,w3,...,wk)' = y 其中：x0=1,w0=b
    :param features_data: 训练数据矩阵
    :param k: 特征数（x向量维度）
    :param labels: 标签向量（y向量）
    :return:
    """
    m = features_data.shape[0]
    print("m=", m)
    n = labels.shape[0]
    print("n=",n)
    x = np.reshape(np.c_[np.ones(m),features_data],[m,k+1])
    y = np.reshape(labels,[n,1])
    return x,y

#1. 特征训练数据
X_train = np.matrix([[1,3,5,2],[3,3,1,2],[1,3,5,2],[3,3,1,2],[3,3,1,2],[3,3,1,2]])
Y_train = np.array([1,2,2,4,3,6])
#特征数量
k = len(X_train[0])

#2. 对训练数据整形，并且X_train,Y_train 必须转换为numpy矩阵或者数组
X_train,Y_train = appendBiasReshape(X_train, Y_train)
m = len(X_train)
n = k+1

#3. X,Y 训练输入站位符
X = tf.placeholder(tf.float32, name="X", shape=[m, n])
Y = tf.placeholder(tf.float32, name="Y")

#4. 定义权重向量
w = tf.Variable(tf.random_normal([n, 1]))

#5. 定义目标函数
Y_hat = tf.matmul(X, w)

#6. 定义损失函数
loss = tf.reduce_mean(tf.square(Y - Y_hat), name="loss")

#7， 选择梯度下降法优化器
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01, name="optimizer").minimize(loss)

#8. 定义变量初始化器
init_op = tf.global_variables_initializer()
total = []


with tf.Session() as sess:
    """ 执行变量初始化 """
    sess.run(init_op)
    """ 保存会话执行图 """
    writer = tf.summary.FileWriter("graphs", sess.graph)
    """ 对训练数据执行100次训练 """
    for i in range(100):
        """ 执行训练：输出每一次的损失函数值，通过feed_dict指定训练参数值"""
        _,l = sess.run([optimizer, loss], feed_dict={X: X_train, Y: Y_train})
        total.append(l)
        print('epoch{0}:loss ={1}'.format(i,l))
    writer.close()
    """ 从sess中获取训练后的权重向量 """
    w_value = sess.run(w)
    print('w_value ={0}, b_value:{1}'.format(w_value,X_train[:, 0]))


plt.plot(total)
plt.show()

TensorBoard 图示：
在控制台输入命令： tensorboard --logdir graphs