希望對你有幫助呀！！💜💜 如有更好理解的思路，歡迎大家留言補充 ~ 一起加油叭 💦
歡迎關(guān)注、訂閱專欄 【深度學(xué)習從 0 到 1】謝謝你的支持！

? 手寫數(shù)字分類: Keras + MNIST 數(shù)據(jù)集

手寫數(shù)字分類任務(wù)

任務(wù)：將手寫數(shù)字的灰度圖像（28像素×28像素）劃分到10個類別中（0~9）

MNIST數(shù)據(jù)集包含60 000張訓(xùn)練圖像和10 000張測試圖像，由美國國家標準與技術(shù)研究院（National Institute of Standards and Technology，即 MNIST 中的NIST）在20世紀80年代收集得到

• 樣本示例如下：(hint: 顯示數(shù)據(jù)集的第一個數(shù)字的代碼：plt.imshow(train_images[0], cmap=plt.cm.binary))

💜步驟一 ： 加載Keras中的MNIST數(shù)據(jù)集

from keras.datasets import mnist 
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()  # 包括4個Numpy數(shù)組# 準備數(shù)據(jù) 
train_images = train_images.reshape((60000, 28 * 28)) 
train_images = train_images.astype('float32') / 255 
test_images  = test_images.reshape((10000, 28 * 28)) 
test_images  = test_images.astype('float32') / 255# 準備標簽
from keras.utils import to_categorical 
train_labels = to_categorical(train_labels) 
test_labels  = to_categorical(test_labels)

💜步驟二 ： 構(gòu)建網(wǎng)絡(luò)架構(gòu) (兩層全連接層為例)

from keras import models 
from keras import layers 
network = models.Sequential() 
network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,))) 
network.add(layers.Dense(10, activation='softmax'))

💜步驟三： 編譯步驟 (optimizer + loss + metrics)

network.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

💜步驟四：訓(xùn)練網(wǎng)絡(luò)

network.fit(train_images, train_labels, epochs=5, batch_size=128)

💜步驟五：測試網(wǎng)絡(luò)

 test_loss, test_acc = network.evaluate(test_images, test_labels) 

---

完整代碼參考：

from keras.datasets import mnist 
from keras import models 
from keras import layers (train_images, train_labels), (test_images, test_labels) = mnist.load_data()  # 包括4個Numpy數(shù)組# 準備數(shù)據(jù) 
train_images = train_images.reshape((60000, 28 * 28)) 
train_images = train_images.astype('float32') / 255 
test_images  = test_images.reshape((10000, 28 * 28)) 
test_images  = test_images.astype('float32') / 255# 準備標簽
from keras.utils import to_categorical 
train_labels = to_categorical(train_labels) 
test_labels  = to_categorical(test_labels)# 構(gòu)建網(wǎng)絡(luò)架構(gòu)
network = models.Sequential() 
network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,))) 
network.add(layers.Dense(10, activation='softmax'))# 編譯步驟
network.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])# 訓(xùn)練網(wǎng)絡(luò)
network.fit(train_images, train_labels, epochs=5, batch_size=128)# 測試網(wǎng)絡(luò)
test_loss, test_acc = network.evaluate(test_images, test_labels) print("Loss: {}, Acc: {}".format(test_loss, test_acc))

----- 結(jié)束后會得到類似如下結(jié)果：

Epoch 1/5
469/469 [==============================] - 2s 5ms/step - loss: 0.2598 - accuracy: 0.9253
Epoch 2/5
469/469 [==============================] - 2s 5ms/step - loss: 0.1041 - accuracy: 0.9692
Epoch 3/5
469/469 [==============================] - 2s 5ms/step - loss: 0.0684 - accuracy: 0.9795
Epoch 4/5
469/469 [==============================] - 2s 5ms/step - loss: 0.0492 - accuracy: 0.9848
Epoch 5/5
469/469 [==============================] - 2s 5ms/step - loss: 0.0367 - accuracy: 0.9892
313/313 [==============================] - 0s 702us/step - loss: 0.0665 - accuracy: 0.9803
Loss: 0.06652633100748062, Acc: 0.9803000092506409

---

參考書籍：Python 深度學(xué)習