tensorflow学习笔记（一） --- 一元线性模型

最近学习tensorflow入门用的资料，觉得讲的很透彻。

在这里给大家分享一下！！！

下面是tensorflow的代码

'''
使用TensorFlow实现一个线性回归算法.
'''
import numpy as np
import matplotlib.pyplot as plt
import os
import tensorflow as tf
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# 产生训练数据集
train_X = np.asarray([3.3,4.4,5.5,6.71,6.93,4.168,9.779,6.182,7.59,2.167,
                         7.042,10.791,5.313,7.997,5.654,9.27,3.1])
train_Y = np.asarray([1.7,2.76,2.09,3.19,1.694,1.573,3.366,2.596,2.53,1.221,
                         2.827,3.465,1.65,2.904,2.42,2.94,1.3])
n_train_samples = train_X.shape[0]
print('训练样本数量: ', n_train_samples)
# 产生测试样本
test_X = np.asarray([6.83, 4.668, 8.9, 7.91, 5.7, 8.7, 3.1, 2.1])
test_Y = np.asarray([1.84, 2.273, 3.2, 2.831, 2.92, 3.24, 1.35, 1.03])
n_test_samples = test_X.shape[0]
print('测试样本数量: ', n_test_samples)

# 展示原始数据分布
plt.plot(train_X, train_Y, 'ro', label='Original Train Points')
plt.plot(test_X, test_Y, 'b*', label='Original Test Points')
plt.legend()
plt.show()

print('~~~~~~~~~~开始设计计算图~~~~~~~~')
# 告诉TensorFlow模型将会被构建在默认的Graph上.
with tf.Graph().as_default():
    # Input: 定义输入节点
    with tf.name_scope('Input'):
        # 计算图输入占位符
        X = tf.placeholder(tf.float32, name='X')
        Y_true = tf.placeholder(tf.float32, name='Y_true')
    # Inference: 定义预测节点
    with tf.name_scope('Inference'):
        # 回归模型的权重和偏置：np.random.randn()返回一个标准正态分布随机数
        W = tf.Variable(np.random.randn(), name="Weight")
        b = tf.Variable(np.random.randn(), name="Bias")
        # inference: 创建一个线性模型：y = wx + b
        Y_pred = tf.add(tf.multiply(X,W), b)
    #Loss: 定义损失节点
    with tf.name_scope('Loss'):
        # Loss = tf.reduce_sum(tf.pow((Y_pred-Y_true), 2))/(2*n_train_samples)
        TrainLoss = tf.reduce_mean(tf.pow((Y_pred - Y_true), 2))/2
    # Train: 定义训练节点
    with tf.name_scope('Train'):
        # Optimizer: 创建一个梯度下降优化器
        Optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
        # Train: 定义训练节点将梯度下降法应用于Loss
        TrainOp = Optimizer.minimize(TrainLoss)
    # Evaluate: 定义评估节点
    with tf.name_scope('Evaluate'):
        # Loss = tf.reduce_sum(tf.pow((Y_pred-Y_true), 2))/(2*n_test_samples)
        EvalLoss = tf.reduce_mean(tf.pow((Y_pred - Y_true), 2)) / 2
    #Initial:添加所有Variable类型的变量的初始化节点
    InitOp = tf.global_variables_initializer()

    print('把计算图写入事件文件，在TensorBoard里面查看')
    writer = tf.summary.FileWriter(logdir='logs', graph=tf.get_default_graph())
    writer.close()

    print('启动会话，开启训练评估模式，让计算图跑起来')
    sess = tf.Session()
    sess.run(InitOp)  #运行初始化节点，完成初始化

    print("不断的迭代训练并测试模型")
    for step in range(1000):
        _, train_loss, train_w, train_b = sess.run([TrainOp, TrainLoss, W, b],
                                                   feed_dict={X: train_X, Y_true: train_Y})
        # 每隔几步训练完之后输出当前模型的损失
        if (step + 1) % 5 == 0:
            print("Step:", '%04d' % (step + 1), "train_loss=", "{:.9f}".format(train_loss),
                  "W=", train_w, "b=", train_b)

        # 每隔几步训练完之后对当前模型进行测试
        if (step + 1) % 5 == 0:
            test_loss, test_w, test_b = sess.run([EvalLoss, W, b],
                                                 feed_dict={X: test_X, Y_true: test_Y})
            print("Step:", '%04d' % (step + 1), "test_loss=", "{:.9f}".format(test_loss),
                  "W=", test_w, "b=", test_b)

    print("训练结束!")
    W, b = sess.run([W, b])
    print("得到的模型参数：", "W=", W, "b=", b,)
    training_loss = sess.run(TrainLoss, feed_dict={X: train_X, Y_true: train_Y})
    print("训练集上的损失：", training_loss)
    test_loss = sess.run(EvalLoss, feed_dict={X: test_X, Y_true: test_Y})
    print("测试集上的损失：", test_loss)
    # 展示拟合曲线
    plt.plot(train_X, train_Y, 'ro', label='Original Train Points')
    plt.plot(test_X, test_Y, 'b*', label='Original Test Points')
    plt.plot(train_X, W * train_X + b, label='Fitted Line')
    plt.legend()
    plt.show()

最后是tensorboard上的计算图以及计算结果

非常好的资源，在这里分享给大家，适合入门的小伙伴~