180204 Keras学习笔记(更新ing)

• ​常规操作​

• ​​将整数标签label进行one-hot转换​​
• ​​保存与加载模型权重​​
• ​​加载mnist数据​​
• ​​加载cifar-100数据​​

• ​​Keras as a simplified interface to TensorFlow​​

常规操作

1.将整数标签label进行one-hot转换

from keras.utils.np_utils import to_categorical
int_labels = np.arange(10)
categorical_labels = to_categorical(int_labels, num_classes=None)

2.保存与加载模型权重

model.save_weights('my_model_weights.h5'，overwrite=True)
model.load_weights('my_model_weights.h5'，overwrite=True)

3.加载mnist数据

from keras.datasets import mnist, cifar100
(X_train, y_train), (X_test, y_test) = mnist.load_data()
print('MNIST training data set label distribution', np.bincount(y_train))
print('test distribution', np.bincount(y_test))

4.加载cifar-100数据

from keras.datasets import mnist, cifar100
(X_train, y_train), (X_test, y_test) = cifar100.load_data(label_mode='fine')
y_train = y_train.ravel()
y_test = y_test.ravel()

Keras as a simplified interface to TensorFlow

​​Keras作为TensorFlow的简化界面​​