我刚开始学习深度学习，尝试将 Keras 的 Sequential API 转换为在 CIFAR10 图像数据集上运行的 Functional API，但遇到了一些困难。我已经转换了模型，除了输入层之外看起来是一样的，但 Sequential 模型的平均准确率约为 70%，而我的 Functional 模型的平均准确率约为 10%。我非常希望能得到一些帮助来弄清楚哪里出了问题。以下是我的 Functional 代码：
import tensorflow as tffrom tensorflow import kerasfrom keras import datasets, layers, modelsfrom keras.models import Model, Input, Sequentialimport matplotlib.pyplot as plt下载并准备数据：
(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()# 将像素值归一化为0到1之间train_images, test_images = train_images / 255.0, test_images / 255.0input_shape = train_images[0,:,:,:].shape创建模型：
input = layers.Input(shape=input_shape)x = layers.Conv2D(32, (3, 3), activation='relu',padding='valid')(input)x = layers.MaxPooling2D((2,2))(x)x = layers.Conv2D(64, (3, 3), activation='relu')(x)x = layers.MaxPooling2D((2,2))(x)x = layers.Conv2D(64, (3, 3), activation='relu')(x)x = layers.Flatten()(x)x = layers.Dense(64, activation='relu')(x)x = layers.Dense(10)(x)model = Model(input, x, name='Functional')编译并训练模型：
model.compile(optimizer='adam', loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=['accuracy'])history = model.fit(train_images, train_labels, epochs=10, validation_data=(test_images, test_labels))这是原始 Sequential CNN 的链接，这是一个 Google Colab 笔记本。我非常希望能得到任何帮助来理解和修复问题。提前感谢您。
回答：
似乎SparseCategoricalCrossentropy损失函数有些问题。
请查看这个链接： https://github.com/tensorflow/tensorflow/issues/38632
以下模型的准确率较高：
import tensorflow as tffrom tensorflow import kerasfrom keras import datasets, layers, modelsfrom keras.models import Model, Input, Sequentialimport matplotlib.pyplot as plt(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()# 将像素值归一化为0到1之间train_images, test_images = train_images / 255.0, test_images / 255.0train_labels, test_labels = tf.keras.utils.to_categorical(train_labels, 10) , tf.keras.utils.to_categorical(test_labels, 10)input_shape = train_images[0,:,:,:].shapeinput = layers.Input(shape=input_shape)x = layers.Conv2D(32, (3, 3), activation='relu',padding='valid')(input)x = layers.MaxPooling2D((2,2))(x)x = layers.Conv2D(64, (3, 3), activation='relu')(x)x = layers.MaxPooling2D((2,2))(x)x = layers.Conv2D(64, (3, 3), activation='relu')(x)x = layers.Flatten()(x)x = layers.Dense(64, activation='relu')(x)x = layers.Dense(10, activation='softmax')(x)model = Model(input, x, name='Functional')model.summary()model.compile(optimizer='adam', loss=loss=tf.keras.losses.CategoricalCrossentropy(), metrics=['accuracy'])history = model.fit(train_images, train_labels, epochs=10, validation_data=(test_images, test_labels))conv2d_16 (Conv2D) (None, 30, 30, 32) 896 _________________________________________________________________max_pooling2d_11 (MaxPooling (None, 15, 15, 32) 0 _________________________________________________________________conv2d_17 (Conv2D) (None, 13, 13, 64) 18496 _________________________________________________________________max_pooling2d_12 (MaxPooling (None, 6, 6, 64) 0 _________________________________________________________________conv2d_18 (Conv2D) (None, 4, 4, 64) 36928 _________________________________________________________________flatten_6 (Flatten) (None, 1024) 0 _________________________________________________________________dense_11 (Dense) (None, 64) 65600 _________________________________________________________________dense_12 (Dense) (None, 10) 650 =================================================================Total params: 122,570Trainable params: 122,570Non-trainable params: 0_________________________________________________________________Train on 50000 samples, validate on 10000 samplesEpoch 1/1050000/50000 [==============================] - 15s 305us/step - loss: 1.4870 - accuracy: 0.4600 - val_loss: 1.2874 - val_accuracy: 0.5488Epoch 2/1050000/50000 [==============================] - 15s 301us/step - loss: 1.1365 - accuracy: 0.5989 - val_loss: 1.0789 - val_accuracy: 0.6191Epoch 3/1050000/50000 [==============================] - 15s 301us/step - loss: 0.9869 - accuracy: 0.6547 - val_loss: 0.9506 - val_accuracy: 0.6700Epoch 4/1050000/50000 [==============================] - 15s 301us/step - loss: 0.8896 - accuracy: 0.6907 - val_loss: 0.9509 - val_accuracy: 0.6695Epoch 5/1050000/50000 [==============================] - 16s 311us/step - loss: 0.8135 - accuracy: 0.7151 - val_loss: 0.8688 - val_accuracy: 0.7046Epoch 6/1050000/50000 [==============================] - 15s 303us/step - loss: 0.7566 - accuracy: 0.7351 - val_loss: 0.8411 - val_accuracy: 0.7141