tensorflow2.0线性回归代码实现 代码里面注释详细

import  tensorflow as tf
import  numpy as np
from    tensorflow import keras
import  os


#基础线性模板
class Regressor(keras.layers.Layer):

def __init__(self):
    super(Regressor, self).__init__()

    # here must specify shape instead of tensor !
    # name here is meanless !
    # [dim_in=13, dim_out=1]
    self.w = self.add_variable('meanless-name-w', [13, 1])
    # [dim_out=1]
    self.b = self.add_variable('meanless-name-b', [1])

    print(self.w.shape, self.b.shape)
    print(type(self.w), tf.is_tensor(self.w), self.w.name)
    print(type(self.b), tf.is_tensor(self.b), self.b.name)


def call(self, x):
    x = tf.matmul(x, self.w) + self.b#c从调用可以看出 x为(64,13), w为(13,1),最终输出x为(64,1)
    return x


def main():

    tf.random.set_seed(22)
    np.random.seed(22)
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
    assert tf.__version__.startswith('2.')


    (x_train, y_train), (x_val, y_val) = keras.datasets.boston_housing.load_data('F:/PycharmProjects/TensorFlow-2.0-Tutorials/boston/boston_housing.npz')
    x_train, x_val = x_train.astype(np.float32), x_val.astype(np.float32)#将x_train和x_val的数据转为float32类型
    print(x_train.shape, y_train.shape, x_val.shape, y_val.shape)# (404, 13) (404,) (102, 13) (102,)
    # Here has two mis-leading issues:
    db_train = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(64)#将x_train,y_train组合 并切分成404个(13,)的矩阵,并将404个矩阵 按照64个为一批 进行分批。
    db_val = tf.data.Dataset.from_tensor_slices((x_val, y_val)).batch(102)#将x_val,y_val组合 并切分成102个(13,)的矩阵,并将102个矩阵 按照102个为一批 进行分批。


    model = Regressor()#
    criteon = keras.losses.MeanSquaredError()#均方误差损失函数
    optimizer = keras.optimizers.Adam(learning_rate=1e-2)#优化函数

    for epoch in range(200):
        for step, (x, y) in enumerate(db_train):#step就是第几批 批数 = 404/64 = 6.3，step的值就是0-6。64个为一批，一个有13个数据，因此x为(64,13)，y为(64,)
            with tf.GradientTape() as tape:
                # 输出：[b=64, 1]
                logits = model(x)
                # 输出[b=64]
                logits = tf.squeeze(logits, axis=1)#从张量形状中移除大小为1的维度，(64,1)移去1后变为(64,)
                # [b] vs [b]
                loss = criteon(y, logits)

            grads = tape.gradient(loss, model.trainable_variables)
            # name = [v.name for v in model.trainable_variables]
            # print("name:", name)#name: ['meanless-name-w:0', 'meanless-name-b:0']
            optimizer.apply_gradients(zip(grads, model.trainable_variables))

        print(epoch, 'loss:', loss.numpy())


        if epoch % 10 == 0:

            for x, y in db_val:
                # [b, 1]
                logits = model(x)
                # [b]
                logits = tf.squeeze(logits, axis=1)
                # [b] vs [b]
                loss = criteon(y, logits)

                print(epoch, 'val loss:', loss.numpy())





if __name__ == '__main__':
    main()