前向传播 | GHBlog

tensorflow基本操作

创建tensor

tf.convert_to_tensor(np.ones([2, 3]))
tf.convert_to_tensor(np.zeros([2, 3]))

tf.convert_to_tensor([1, 2])
tf.convert_to_tensor([[1], [2]])

tf.ones([2, 2])
tf.ones([1, 4, 2])
tf.ones([2, 4, 2])
tf.ones([3, 4, 2])

tf.fill([2, 3], 6)

tf.random.normal([2, 2], mean=1, stddev=1)  # 正态分布
tf.random.uniform([2, 2], minval=0, maxval=1)  # 均匀分布

tf.ones([2, 5, 5, 3])

前向传播实战

step1 导入模块

1 2	import tensorflow as tf from tensorflow.keras import datasets

step2 导入数据集

1	(x_train, y_train), _ = datasets.mnist.load_data()

step3 处理数据集

x_train = tf.convert_to_tensor(x_train, dtype=tf.float32) / 255
y_train = tf.convert_to_tensor(y_train, dtype=tf.int32)
# print(tf.reduce_max(x_train))

train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(128)
train_iter = iter(train_db)
sample = next(train_iter)

step4 设置参数

# [b, 784] -> [784, 256] -> [256, 10]
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
b1 = tf.Variable(tf.zeros(256))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 = tf.Variable(tf.zeros(128))
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros(10))

step5训练模型

for epoch in range(10):  # iterate dataset
    for step, (x_train, y_train) in enumerate(train_db):    # for every batch  # x.shape: [128, 28, 28] y.shape[128]
        x_train = tf.reshape(x_train, [-1, 28*28])  # x.shape: [b, 784]
        with tf.GradientTape() as tape:
            h1 = x_train @ w1 + b1  # h1.shape: [b, 256]
            h1 = tf.nn.relu(h1)
            h2 = h1 @ w2 + b2  # h2.shape: [b, 128]
            h2 = tf.nn.relu(h2)
            out = h2 @ w3 + b3  # out.shape: [b, 10]

            # compute loss
            y_hot = tf.one_hot(y_train, depth=10)
            loss = tf.square(y_hot - out)
            loss = tf.reduce_mean(loss)

        # compute gradients
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
        # update gradients
        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])

        # w1 = tf.Variable(w1 - lr * grads[0])
        # b1 = tf.Variable(b1 - lr * grads[1])
        # w2 = tf.Variable(w2 - lr * grads[2])
        # b2 = tf.Variable(b2 - lr * grads[3])
        # w3 = tf.Variable(w3 - lr * grads[4])
        # b3 = tf.Variable(b3 - lr * grads[5])

        if step % 100 == 0:
            print('epoch:', epoch, 'step:', step, 'current loss:', float(loss))

完整代码

import tensorflow as tf
from tensorflow.keras import datasets

lr = 0.001

(x_train, y_train), _ = datasets.mnist.load_data()
x_train = tf.convert_to_tensor(x_train, dtype=tf.float32) / 255
y_train = tf.convert_to_tensor(y_train, dtype=tf.int32)
# print(tf.reduce_max(x_train))

train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(128)
train_iter = iter(train_db)
sample = next(train_iter)
# print('batch:', sample[0].shape, sample[1].shape)

# [b, 784] -> [784, 256] -> [256, 10]
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
b1 = tf.Variable(tf.zeros(256))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 = tf.Variable(tf.zeros(128))
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros(10))

for epoch in range(10):  # iterate dataset
    for step, (x_train, y_train) in enumerate(train_db):    # for every batch  # x.shape: [128, 28, 28] y.shape[128]
        x_train = tf.reshape(x_train, [-1, 28*28])  # x.shape: [b, 784]
        with tf.GradientTape() as tape:
            h1 = x_train @ w1 + b1  # h1.shape: [b, 256]
            h1 = tf.nn.relu(h1)
            h2 = h1 @ w2 + b2  # h2.shape: [b, 128]
            h2 = tf.nn.relu(h2)
            out = h2 @ w3 + b3  # out.shape: [b, 10]

            # compute loss
            y_hot = tf.one_hot(y_train, depth=10)
            loss = tf.square(y_hot - out)
            loss = tf.reduce_mean(loss)

        # compute gradients
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
        # update gradients
        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])

        # w1 = tf.Variable(w1 - lr * grads[0])
        # b1 = tf.Variable(b1 - lr * grads[1])
        # w2 = tf.Variable(w2 - lr * grads[2])
        # b2 = tf.Variable(b2 - lr * grads[3])
        # w3 = tf.Variable(w3 - lr * grads[4])
        # b3 = tf.Variable(b3 - lr * grads[5])

        if step % 100 == 0:
            print('epoch:', epoch, 'step:', step, 'current loss:', float(loss))