// 未分类
【调包侠的机器学习】图片分类
发布 2022-08-23 · 永久链接
```python
%matplotlib inline
import tensorflow as tf
!pip install d2l
from d2l import tensorflow as d2l
```
```python
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.fashion_mnist.load_data()
```
```python
# y_train
# d2l.show_images(tf.constant(x_train[:10]), 1, 10)
x_train.shape, y_train.shape
```
((60000, 28, 28), (60000,))
```python
# one hot y, 10 classes
y_train_hot = tf.one_hot(y_train[:], 10)
y_train_hot.shape
```
TensorShape([60000, 10])
```python
x_train_shaped = tf.reshape(x_train, [60000, -1]) / 255
x_train_shaped.shape, x_train_shaped[:1]
```
(TensorShape([60000, 784]),
)
```python
net = tf.keras.Sequential([
tf.keras.layers.Dense(units=10, input_dim=784, activation=tf.keras.activations.softmax)
])
sgd = tf.keras.optimizers.SGD(learning_rate=0.01)
net.compile(optimizer=sgd, loss='mse', metrics=['accuracy'])
```
```python
history = net.fit(x_train_shaped, y_train_hot, batch_size=1000, epochs=500, validation_split=0.2)
net.summary()
```
Epoch 1/500
48/48 [==============================] - 0s 4ms/step - loss: 0.0957 - accuracy: 0.0224 - val_loss: 0.0950 - val_accuracy: 0.0231
Epoch 500/500
48/48 [==============================] - 0s 2ms/step - loss: 0.0347 - accuracy: 0.7809 - val_loss: 0.0346 - val_accuracy: 0.7787
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_3 (Dense) (None, 10) 7850
=================================================================
Total params: 7,850
Trainable params: 7,850
Non-trainable params: 0
_________________________________________________________________
```python
print(x_test.shape)
x_test_shaped = tf.reshape(x_test, [10000, -1]) / 255
y_test_hot = tf.one_hot(y_test, 10)
x_test_shaped.shape, y_test_hot.shape
```
(10000, 28, 28)
(TensorShape([10000, 784]), TensorShape([10000, 10]))
```python
net.evaluate(x_test_shaped, y_test_hot)
```
313/313 [==============================] - 1s 2ms/step - loss: 0.0357 - accuracy: 0.7704
[0.03569139540195465, 0.7703999876976013]
```python
# !pip install matplotlib
import matplotlib.pyplot as plt
print("keys: ", history.history.keys())
plt.plot(history.history["loss"], label="loss")
plt.plot(history.history["accuracy"], label="accuracy")
plt.plot(history.history["val_loss"], label="val_loss")
plt.plot(history.history["val_accuracy"], label="val_accuracy")
plt.legend()
plt.show()
```
keys: dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])
```python
py = net.predict(x_test_shaped)
py.argmax(axis=1).shape
```
313/313 [==============================] - 0s 866us/step
(10000,)