# -*- coding: utf-8 -*-
import numpy as np
import matplotlib.pyplot as plt
class SimpleDataReader(object):
    def __init__(self, data_file):
        self.train_file_name = data_file
        self.num_train = 0
        self.XTrain = None
        self.YTrain = None
     # read data from file
    def ReadData(self):
        data = np.load(self.train_file_name)
        self.XTrain = data["data"]
        self.YTrain = data["label"]
        self.num_train = self.XTrain.shape[0]
        #end if
    # get batch training data
    def GetSingleTrainSample(self, iteration):
        x = self.XTrain[iteration]
        y = self.YTrain[iteration]
        return x, y
    def GetWholeTrainSamples(self):
        return self.XTrain, self.YTrain
class NeuralNet(object):
    def __init__(self, eta):
        self.eta = eta
        self.w = 0
        self.b = 0
    def __forward(self, x):
        z = x * self.w + self.b
        return z
    def __backward(self, x,y,z):
        dz = z - y
        db = dz
        dw = x * dz
        return dw, db
    def __update(self, dw, db):
        self.w = self.w - self.eta * dw
        self.b = self.b - self.eta * db
    def train(self, dataReader):
        for i in range(dataReader.num_train):
            # get x and y value for one sample
            x,y = dataReader.GetSingleTrainSample(i)
            # get z from x,y
            z = self.__forward(x)
            # calculate gradient of w and b
            dw, db = self.__backward(x, y, z)
            # update w,b
            self.__update(dw, db)
            # end for
    def inference(self, x):
        return self.__forward(x)
if __name__ == '__main__':
    # read data
    sdr = SimpleDataReader('ch04.npz')
    sdr.ReadData()
    # create net
    eta = 0.1
    net = NeuralNet(eta)
    net.train(sdr)
    # result
    print("w=%f,b=%f" %(net.w, net.b))
    # 繪圖部分
    trainX,trainY = sdr.GetWholeTrainSamples()
    fig = plt.figure()
    ax = fig.add_subplot(111)
    # 繪制散點(diǎn)圖
    ax.scatter(trainX,trainY)
    # 繪制線性回歸
    x = np.arange(0, 1, 0.01)
    f = np.vectorize(net.inference, excluded=['x'])
    plt.plot(x,f(x),color='red')
    # 顯示圖表
    plt.show()

tf.matmul(x, w)

tf.square(error)

tf.reduce_mean(error)

tf.train.GradientDescentOptimizer(learning_rate)

import tensorflow as tf
import os
def linear_regression():
? ? """
? ? 自實(shí)現(xiàn)線性回歸
? ? :return: None
? ? """
? ? # 1）準(zhǔn)備好數(shù)據(jù)集：y = 0.8x + 0.7 100個樣本
? ? # 特征值X, 目標(biāo)值y_true
? ? X = tf.random_normal(shape=(100, 1), mean=2, stddev=2)
? ? # y_true [100, 1]
? ? # 矩陣運(yùn)算 X（100， 1）* （1, 1）= y_true(100, 1)
? ? y_true = tf.matmul(X, [[0.8]]) + 0.7
? ? # 2）建立線性模型：
? ? # y = W·X + b，目標(biāo)：求出權(quán)重W和偏置b
? ? # 3）隨機(jī)初始化W1和b1
? ? weights = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)))
? ? bias = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)))
? ? y_predict = tf.matmul(X, weights) + bias
? ? # 4）確定損失函數(shù)（預(yù)測值與真實(shí)值之間的誤差）-均方誤差
? ? error = tf.reduce_mean(tf.square(y_predict - y_true))
? ? # 5）梯度下降優(yōu)化損失：需要指定學(xué)習(xí)率（超參數(shù)）
? ? # W2 = W1 - 學(xué)習(xí)率*(方向)
? ? # b2 = b1 - 學(xué)習(xí)率*(方向)
? ? optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(error)
? ? # 初始化變量
? ? init = tf.global_variables_initializer()
? ? # 開啟會話進(jìn)行訓(xùn)練
? ? with tf.Session() as sess:
? ? ? ? # 運(yùn)行初始化變量Op
? ? ? ? sess.run(init)
? ? ? ? print("隨機(jī)初始化的權(quán)重為%f， 偏置為%f" % (weights.eval(), bias.eval()))
? ? ? ? # 訓(xùn)練模型
? ? ? ? for i in range(100):
? ? ? ? ? ? sess.run(optimizer)
? ? ? ? ? ? print("第%d步的誤差為%f，權(quán)重為%f， 偏置為%f" % (i, error.eval(), weights.eval(), bias.eval()))
? ? return None

weight = tf.Variable(tf.random_normal([1, 1], mean=0.0, stddev=1.0), name="weights", trainable=False)

with tf.variable_scope("lr_model"):

def linear_regression():
    # 1）準(zhǔn)備好數(shù)據(jù)集：y = 0.8x + 0.7 100個樣本
    # 特征值X, 目標(biāo)值y_true
    with tf.variable_scope("original_data"):
        X = tf.random_normal(shape=(100, 1), mean=2, stddev=2, name="original_data_x")
        # y_true [100, 1]
        # 矩陣運(yùn)算 X（100， 1）* （1, 1）= y_true(100, 1)
        y_true = tf.matmul(X, [[0.8]], name="original_matmul") + 0.7
    # 2）建立線性模型：
    # y = W·X + b，目標(biāo)：求出權(quán)重W和偏置b
    # 3）隨機(jī)初始化W1和b1
    with tf.variable_scope("linear_model"):
        weights = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)), name="weights")
        bias = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)), name="bias")
        y_predict = tf.matmul(X, weights, name="model_matmul") + bias
    # 4）確定損失函數(shù)（預(yù)測值與真實(shí)值之間的誤差）-均方誤差
    with tf.variable_scope("loss"):
        error = tf.reduce_mean(tf.square(y_predict - y_true), name="error_op")
    # 5）梯度下降優(yōu)化損失：需要指定學(xué)習(xí)率（超參數(shù)）
    # W2 = W1 - 學(xué)習(xí)率*(方向)
    # b2 = b1 - 學(xué)習(xí)率*(方向)
    with tf.variable_scope("gd_optimizer"):
        optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01, name="optimizer").minimize(error)
    # 2）收集變量
    tf.summary.scalar("error", error)
    tf.summary.histogram("weights", weights)
    tf.summary.histogram("bias", bias)
    # 3）合并變量
    merge = tf.summary.merge_all()
    # 初始化變量
    init = tf.global_variables_initializer()
    # 開啟會話進(jìn)行訓(xùn)練
    with tf.Session() as sess:
        # 運(yùn)行初始化變量Op
        sess.run(init)
        print("隨機(jī)初始化的權(quán)重為%f， 偏置為%f" % (weights.eval(), bias.eval()))
        # 1）創(chuàng)建事件文件
        file_writer = tf.summary.FileWriter(logdir="./summary", graph=sess.graph)
        # 訓(xùn)練模型
        for i in range(100):
            sess.run(optimizer)
            print("第%d步的誤差為%f，權(quán)重為%f， 偏置為%f" % (i, error.eval(), weights.eval(), bias.eval()))
            # 4）運(yùn)行合并變量op
            summary = sess.run(merge)
            file_writer.add_summary(summary, i)
    return None

tf.train.Saver(var_list=None,max_to_keep=5)

saver.save(sess, '/tmp/ckpt/test/myregression.ckpt')
saver.restore(sess, '/tmp/ckpt/test/myregression.ckpt')

checkpoint = tf.train.latest_checkpoint("./tmp/model/")
saver.restore(sess, checkpoint)

# 定義一些常用的命令行參數(shù)
# 訓(xùn)練步數(shù)
tf.app.flags.DEFINE_integer("max_step", 0, "訓(xùn)練模型的步數(shù)")
# 定義模型的路徑
tf.app.flags.DEFINE_string("model_dir", " ", "模型保存的路徑+模型名字")
# 定義獲取命令行參數(shù)
FLAGS = tf.app.flags.FLAGS
# 開啟訓(xùn)練
# 訓(xùn)練的步數(shù)（依據(jù)模型大小而定）
for i in range(FLAGS.max_step):
     sess.run(train_op)

import tensorflow as tf
import os
tf.app.flags.DEFINE_string("model_path", "./linear_regression/", "模型保存的路徑和文件名")
FLAGS = tf.app.flags.FLAGS
def linear_regression():
? ? # 1）準(zhǔn)備好數(shù)據(jù)集：y = 0.8x + 0.7 100個樣本
? ? # 特征值X, 目標(biāo)值y_true
? ? with tf.variable_scope("original_data"):
? ? ? ? X = tf.random_normal(shape=(100, 1), mean=2, stddev=2, name="original_data_x")
? ? ? ? # y_true [100, 1]
? ? ? ? # 矩陣運(yùn)算 X（100， 1）* （1, 1）= y_true(100, 1)
? ? ? ? y_true = tf.matmul(X, [[0.8]], name="original_matmul") + 0.7
? ? # 2）建立線性模型：
? ? # y = W·X + b，目標(biāo)：求出權(quán)重W和偏置b
? ? # 3）隨機(jī)初始化W1和b1
? ? with tf.variable_scope("linear_model"):
? ? ? ? weights = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)), name="weights")
? ? ? ? bias = tf.Variable(initial_value=tf.random_normal(shape=(1, 1)), name="bias")
? ? ? ? y_predict = tf.matmul(X, weights, name="model_matmul") + bias
? ? # 4）確定損失函數(shù)（預(yù)測值與真實(shí)值之間的誤差）-均方誤差
? ? with tf.variable_scope("loss"):
? ? ? ? error = tf.reduce_mean(tf.square(y_predict - y_true), name="error_op")
? ? # 5）梯度下降優(yōu)化損失：需要指定學(xué)習(xí)率（超參數(shù)）
? ? # W2 = W1 - 學(xué)習(xí)率*(方向)
? ? # b2 = b1 - 學(xué)習(xí)率*(方向)
? ? with tf.variable_scope("gd_optimizer"):
? ? ? ? optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01, name="optimizer").minimize(error)
? ? # 2）收集變量
? ? tf.summary.scalar("error", error)
? ? tf.summary.histogram("weights", weights)
? ? tf.summary.histogram("bias", bias)
? ? # 3）合并變量
? ? merge = tf.summary.merge_all()
? ? # 初始化變量
? ? init = tf.global_variables_initializer()
? ? # 創(chuàng)建saver對象
? ? saver = tf.train.Saver()
? ? # 開啟會話進(jìn)行訓(xùn)練
? ? with tf.Session() as sess:
? ? ? ? # 運(yùn)行初始化變量Op
? ? ? ? sess.run(init)
? ? ? ? # 未經(jīng)訓(xùn)練的權(quán)重和偏置
? ? ? ? print("隨機(jī)初始化的權(quán)重為%f， 偏置為%f" % (weights.eval(), bias.eval()))
? ? ? ? # 當(dāng)存在checkpoint文件，就加載模型
? ? ? ? if os.path.exists("./linear_regression/checkpoint"):
? ? ? ? ? ? saver.restore(sess, FLAGS.model_path)
? ? ? ? # 1）創(chuàng)建事件文件
? ? ? ? file_writer = tf.summary.FileWriter(logdir="./summary", graph=sess.graph)
? ? ? ? # 訓(xùn)練模型
? ? ? ? for i in range(100):
? ? ? ? ? ? sess.run(optimizer)
? ? ? ? ? ? print("第%d步的誤差為%f，權(quán)重為%f， 偏置為%f" % (i, error.eval(), weights.eval(), bias.eval()))
? ? ? ? ? ? # 4）運(yùn)行合并變量op
? ? ? ? ? ? summary = sess.run(merge)
? ? ? ? ? ? file_writer.add_summary(summary, i)
? ? ? ? ? ? if i % 10 == 0:
? ? ? ? ? ? ? ? saver.save(sess, FLAGS.model_path)
? ? return None
def main(argv):
? ? print("這是main函數(shù)")
? ? print(argv)
? ? print(FLAGS.model_path)
? ? linear_regression()
if __name__ == "__main__":
? ? tf.app.run()