训练集、验证集和测试集随机划分实现代码

VALIDATION_PERCENTAGE = 10
TEST_PERCENTAGE = 10

validation_images=[]
validation_labels=[]
testing_images=[]
testing_labels=[]
training_images=[]
training_labels=[]

chance = np.random.randint(100)
if chance < VALIDATION_PERCENTAGE:
    validation_images.append(image_value)
    validation_labels.append(image_label)
elif chance < (TEST_PERCENTAGE + VALIDATION_PERCENTAGE):
    testing_images.append(image_value)
    testing_labels.append(image_label)
else:
    training_images.append(image_value)
    training_labels.append(image_label)

训练集就像题海战术，验证集就像模拟测验，测试集就像高考。

验证集通常在训练过程中验证，如每一个epoch，目的在于训练过程实时观察准确率。

验证集可以在训练集中挑选和重复，测试集一般不和训练集重复。

方法二

from sklearn.model_selection import train_test_split
train_images, validation_images = train_test_split(all_images, train_size=0.8, test_size=0.2)