Tensorflow 2.0 实现汽车效能指标 MPG 预测

    终究还是要学习Tensorflow 2.0 的，发现确实2.0比1.0版本的好太多了，感觉精简了些，同时训练的细节要自己实现，这比keras更锻炼自己一些，keras封装太厉害啦。而且可能是我学习的tf2.0少，学到目前还都不需要构建数据集的生成器，这点比keras, pytorch又省了一波。
    下面附上所用的数据集，来自UCI：MPG数据集链接。
    我只实现了数据集的准备，训练部分，测试的话没啥意思，重点是学习Tensorflow 2.0 的语法。

# -*- coding: utf- 8 -*-

'''
Author: Unlabel
Date: 2020-10-09 09:50:44
LastEditTime: 2020-10-09 16:20:06
Description: 利用全连接网络模型来完成汽车的效能指标 MPG(Mile Per Gallon，每加仑
燃油英里数)的预测问题实战。
FilePath: /tf2-learning/Auto_MPG.py
'''

import os
import tensorflow as tf
import pandas as pd


def norm(x):
    """标准化数据"""
    return (x - train_stats['mean']) / train_stats['std']


class Net(tf.keras.Model):
    def __init__(self):
        super(Net, self).__init__()
        self.fc1 = tf.keras.layers.Dense(64, activation='relu', input_dim=9)
        self.fc2 = tf.keras.layers.Dense(64, activation='relu')
        self.fc3 = tf.keras.layers.Dense(1, activation=None)

    def call(self, inputs):
        x = self.fc1(inputs)
        x = self.fc2(x)
        x = self.fc3(x)
        return x


def train(train_db, epochs):
    for epoch in range(epochs):
        for step, (x, y) in enumerate(train_db):
            with tf.GradientTape() as tape:
                y_hat = model(x)
                loss = tf.reduce_mean(tf.keras.losses.mse(y, y_hat))
                grads = tape.gradient(loss, model.trainable_variables)
                optimizer.apply_gradients(zip(grads, model.trainable_variables))
            print("epoch: %d / step: %d --------- loss: %f" % (epoch, step, loss))


if __name__ == "__main__":
    data_path = os.path.join(os.getcwd(), 'data', 'auto-mpg.data')
    column_names = ['MPG', 'Cylinders', 'Displacement', 'Horsepower', 'Weight', 'Acceleration', 'Model Year', 'Origin']
    row_dataset = pd.read_csv(data_path, names=column_names, na_values="?", comment="\t", sep=" ", skipinitialspace=True)
    dataset = row_dataset.copy()
    # 统计空白数据
    # print(dataset.isna().sum())
    # 清除空白项
    dataset = dataset.dropna()
    origin = dataset.pop("Origin")
    dataset["USA"] = (origin == 1) * 1.0
    dataset["Europe"] = (origin == 2) * 1.0
    dataset["Japan"] = (origin == 3) * 1.0
    # 划分训练集和测试集
    train_data = dataset.sample(frac=0.8, random_state=0)
    test_data = dataset.drop(train_data.index)
    # 移除MPA能效这一列为真实标签Y
    # pop仅可以删除列， drop不仅可以删除列还可以删除行
    train_label = train_data.pop("MPG")
    test_label = test_data.pop("MPG")
    # describe()方法用来展示数据的一些描述性统计信息，例如：数量、mean、std等等
    train_stats = train_data.describe()
    train_stats = train_stats.transpose()
    normed_train_data = norm(train_data)
    normed_test_data = norm(test_data)
    train_db = tf.data.Dataset.from_tensor_slices((normed_train_data.values, train_label.values))
    train_db = train_db.shuffle(100).batch(16)

    # 构建模型并训练
    model = Net()
    # 32 为设置的 batch-size, 即shuflle().batch(batzh-size)
    model.build(input_shape=(16, 9))
    # model.summary()
    optimizer = tf.keras.optimizers.Adam(learning_rate=0.01)
    train_epochs = 200
    train(train_db, train_epochs)

    欢迎大家批评指正，谢谢大家！