import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics
# datasets :用于数据集管理 layers.Dense:用于构建全连接层, optimizers:优化器 metics:测试的度量器
def preprocess(x, y):
x = tf.cast(x, dtype=tf.float32) / 255 # tf.convert_to_tensor 用于将不同数据转化为张量 可以是numpy数据 python数据 tensor数据
y = tf.cast(y, dtype=tf.int32)
return x, y
(x, y), (x_test, y_test) = datasets.fashion_mnist.load_data() # 加载数据集
print(x.shape, y.shape)
batchsz = 128
db = tf.data.Dataset.from_tensor_slices((x, y)) # 构造数据集
db = db.map(preprocess).shuffle(10000).batch(batchsz) # 数据预处理 这里是传入一个函数,而不是一个函数的调用
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test)) # 构造数据集
db_test = db_test.map(preprocess).batch(batchsz) # 测试集不需要shuffle
db_iter = iter(db) # iter函数用来生成迭代器
sample = next(db_iter)
print('batch:',sample[0].shape, sample[1].shape)
model = Sequential([
layers.Dense(256, activation=tf.nn.relu), # [b, 784] => [b, 256]
layers.Dense(128, activation=tf.nn.relu), # [b, 256] => [b, 128]
layers.Dense(64, activation=tf.nn.relu), # [b, 128] => [b, 64]
layers.Dense(32, activation=tf.nn.relu), # [b, 64] => [b, 32]
layers.Dense(10) # [b, 32] => [b, 10]
])
model.build(input_shape=[None, 28*28]) # 建立模型,传入数据
model.summary() # 调试的功能,作用是打印网络结构
optimizer = optimizers.Adam(lr=1e-3)
def main():
for epoch in range(30): # 数据集训练30遍
for step, (x, y) in enumerate(db):
# x: [b, 28*28]
# y:[b]
x = tf.reshape(x, [-1, 28*28])
y_one_hot = tf.one_hot(y, depth=10)
with tf.GradientTape() as tape:
logits = model(x) # 把没有加激活函数的输出值称作logits。
loss = tf.reduce_mean(tf.losses.MSE(y_one_hot, logits)) # 均方差损失函数。
loss2 = tf.losses.categorical_crossentropy(y_one_hot, logits, from_logits=True) # 交叉熵损失函数,直接用logits进行运算,一定要设置from_logits=True。这里tf进行了封装,内部会转化为softmax并有一些优化,效果上,比手动算出softmax,然后计算loss,数值更稳定。
loss2 = tf.reduce_mean(loss2) # 由于上一步计算的loss2是tensor,这里要做一下求平均值
grads = tape.gradient(loss2, model.trainable_variables) # model.trainable_variables返回变量列表,不需要我们再额外管理变量
optimizer.apply_gradients(zip(grads, model.trainable_variables)) # 反向传播计算。参数中 梯度(grads)和变量要一一对应,所以用了zip()
if step % 100 == 0:
print(epoch, step, 'loss:', float(loss2), float(loss))
# test
total_correct = 0
total_num = 0
for step, (x_test, y_test) in enumerate(db_test):
# x: [b, 28*28]
# y:[b] 测试情况下,y不需要做one_hot
x = tf.reshape(x, [-1, 28*28])
# [b, 28*28] => [b, 10]
logits = model(x)
# logits => prob [b, 10]
prob = tf.nn.softmax(logits, axis=1)
# [b, 10] => [b]
pred = tf.argmax(prob, axis=1)
pred = tf.cast(pred, tf.int32)
# pred [b]
# y [b]
correct = tf.equal(pred, y)
correct = tf.reduce_sum(tf.cast(correct, dtype=tf.int32))
total_correct += int(correct)
total_num += x.shape[0]
acc = total_correct / total_num
print(epoch, 'acc:', acc)
if __name__ == '__main__':
main()