import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow.keras.datasets as datasets

plt.rcParams['font.size'] = 16
plt.rcParams['font.family'] = ['STKaiti']
plt.rcParams['axes.unicode_minus'] = False


def load_data():
    # 加载 MNIST 数据集
    (x, y), (x_val, y_val) = datasets.mnist.load_data()
    # 转换为浮点张量， 并缩放到-1~1
    x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.
    # 转换为整形张量
    y = tf.convert_to_tensor(y, dtype=tf.int32)
    # one-hot 编码
    y = tf.one_hot(y, depth=10)

    # 改变视图， [b, 28, 28] => [b, 28*28]
    x = tf.reshape(x, (-1, 28 * 28))

    # 构建数据集对象
    train_dataset = tf.data.Dataset.from_tensor_slices((x, y))
    # 批量训练
    train_dataset = train_dataset.batch(200)
    return train_dataset


def init_paramaters():
    # 每层的张量都需要被优化，故使用 Variable 类型，并使用截断的正太分布初始化权值张量
    # 偏置向量初始化为 0 即可
    # 第一层的参数
    w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
    b1 = tf.Variable(tf.zeros([256]))
    # 第二层的参数
    w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
    b2 = tf.Variable(tf.zeros([128]))
    # 第三层的参数
    w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
    b3 = tf.Variable(tf.zeros([10]))
    return w1, b1, w2, b2, w3, b3


def train_epoch(epoch, train_dataset, w1, b1, w2, b2, w3, b3, lr=0.001):
    for step, (x, y) in enumerate(train_dataset):
        with tf.GradientTape() as tape:
            # 第一层计算， [b, 784]@[784, 256] + [256] => [b, 256] + [256] => [b,256] + [b, 256]
            h1 = x @ w1 + tf.broadcast_to(b1, (x.shape[0], 256))
            h1 = tf.nn.relu(h1)  # 通过激活函数

            # 第二层计算， [b, 256] => [b, 128]
            h2 = h1 @ w2 + b2
            h2 = tf.nn.relu(h2)
            # 输出层计算， [b, 128] => [b, 10]
            out = h2 @ w3 + b3

            # 计算网络输出与标签之间的均方差， mse = mean(sum(y-out)^2)
            # [b, 10]
            loss = tf.square(y - out)
            # 误差标量， mean: scalar
            loss = tf.reduce_mean(loss)

            # 自动梯度，需要求梯度的张量有[w1, b1, w2, b2, w3, b3]
            grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])

        # 梯度更新， assign_sub 将当前值减去参数值，原地更新
        w1.assign_sub(lr * grads[0])
        b1.assign_sub(lr * grads[1])
        w2.assign_sub(lr * grads[2])
        b2.assign_sub(lr * grads[3])
        w3.assign_sub(lr * grads[4])
        b3.assign_sub(lr * grads[5])

        if step % 100 == 0:
            print(epoch, step, 'loss:', loss.numpy())

    return loss.numpy()


def train(epochs):
    losses = []
    train_dataset = load_data()
    w1, b1, w2, b2, w3, b3 = init_paramaters()
    for epoch in range(epochs):
        loss = train_epoch(epoch, train_dataset, w1, b1, w2, b2, w3, b3, lr=0.001)
        losses.append(loss)

    x = [i for i in range(0, epochs)]
    # 绘制曲线
    plt.plot(x, losses, color='blue', marker='s', label='训练')
    plt.xlabel('Epoch')
    plt.ylabel('MSE')
    plt.legend()
    plt.savefig('MNIST数据集的前向传播训练误差曲线.png')
    plt.close()


if __name__ == '__main__':
    train(epochs=20)