### 1、数据集 数据集使用的是Fashion mnist, 主要是衣服鞋子的图片。 ### 2、代码讲解 **datasets：自动下载导入数据包的库** **layers：调用全连接库** **optimizers：参数更新步长，learning rate** **Sequential: 串连接容器** **metrics: 指标（Accuracy，Mean）** ```py import tensorflow as tf from tensorflow import keras from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # 定义一个数据预处理函数 def preprocess(x, y): x = tf.cast(x, dtype=tf.float32) / 255. y = tf.cast(y, dtype=tf.int32) return x,y # 下载fashion mnist 数据 (x, y), (x_test, y_test) = datasets.fashion_mnist.load_data() print(x.shape, y.shape) # 设置批量操作的数据大小，每个step处理的数据 batchsz = 128 # 把x, y,设置成可以批量抽取，shuffle：打乱数据 db = tf.data.Dataset.from_tensor_slices((x,y)) db = db.map(preprocess).shuffle(10000).batch(batchsz) # 同理：x_test, y_test 设置成批量抽取 db_test = tf.data.Dataset.from_tensor_slices((x_test,y_test)) db_test = db_test.map(preprocess).batch(batchsz) # 创建一个迭代器，抽取数据 db_iter = iter(db) # 抽取一个数据看一下格式 sample = next(db_iter) print('batch:', sample[0].shape, sample[1].shape) # 创建串联的全连接层模型，激活函数用relu，把维度由784变为10，因为图片只有10类 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], 330 = 32*10 + 10 ]) # 构建模型 model.build(input_shape=[None, 28*28]) # 查看模型的各种数据详情 model.summary() # w = w - lr*grad，设置更新步长 optimizer = optimizers.Adam(lr=1e-3) # 主运行方法 def main(): # 循环30次数据（一批数据重复训练） for epoch in range(30): # 批量抽取数据，每个step数据量为：128 for step, (x,y) in enumerate(db): # x: [b, 28, 28] => [b, 784] # y: [b] x = tf.reshape(x, [-1, 28*28]) # 求导 with tf.GradientTape() as tape: # [b, 784] => [b, 10] logits = model(x) y_onehot = tf.one_hot(y, depth=10) # 计算loss mse(目前没使用这种loss) loss_mse = tf.reduce_mean(tf.losses.MSE(y_onehot, logits)) # 计算 loss categorical_crossentropy loss_ce = tf.losses.categorical_crossentropy(y_onehot, logits, from_logits=True) loss_ce = tf.reduce_mean(loss_ce) # 得到每个参数的导数方向 grads = tape.gradient(loss_ce, model.trainable_variables) # 按照导数和更新步长，自动更新参数 optimizer.apply_gradients(zip(grads, model.trainable_variables)) if step % 100 == 0: print(epoch, step, 'loss:', float(loss_ce), float(loss_mse)) # test，计算每个epoch的accury total_correct = 0 total_num = 0 for x, y in db_test: # x: [b, 28, 28] => [b, 784] # y: [b] x = tf.reshape(x, [-1, 28*28]) # [b, 10] logits = model(x) # logits => prob, [b, 10]，求得概率 prob = tf.nn.softmax(logits, axis=1) # [b, 10] => [b], int64，求得概率的最大下标 pred = tf.argmax(prob, axis=1) pred = tf.cast(pred, dtype=tf.int32) # correct: [b], True: equal, False: not equal correct = tf.equal(pred, y) # 计算1（True）的个数 correct = tf.reduce_sum(tf.cast(correct, dtype=tf.int32)) # 批量的求和要汇总 total_correct += int(correct) total_num += x.shape[0] # 计算每个epoch的accury acc = total_correct / total_num print(epoch, 'test acc:', acc) if __name__ == '__main__': main() ```