|
|
|
|
|
from datetime import datetime
|
|
|
|
|
|
import tensorflow as tf
|
|
|
|
|
|
from tensorflow.keras.datasets import mnist
|
|
|
|
|
|
from tensorflow.keras.utils import to_categorical
|
|
|
|
|
|
from tensorflow.keras.callbacks import LearningRateScheduler, EarlyStopping, TensorBoard
|
|
|
|
|
|
from model_core import LeNet5Custom
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# 打印模型框架基本信息
|
|
|
|
|
|
print(tf.sysconfig.get_build_info())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def load_and_preprocess_data():
|
|
|
|
|
|
# 加载MNIST数据集
|
|
|
|
|
|
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
|
|
|
|
|
|
|
|
|
|
|
|
# 反转像素值：255 - x
|
|
|
|
|
|
train_images = 255 - train_images
|
|
|
|
|
|
test_images = 255 - test_images
|
|
|
|
|
|
|
|
|
|
|
|
# 归一化像素值到 [0, 1] 区间
|
|
|
|
|
|
train_images = train_images.astype("float32") / 255
|
|
|
|
|
|
test_images = test_images.astype("float32") / 255
|
|
|
|
|
|
|
|
|
|
|
|
# 由于 MNIST 的图像是灰度图像，需要增加一个颜色通道
|
|
|
|
|
|
train_images = train_images.reshape(-1, 28, 28, 1)
|
|
|
|
|
|
test_images = test_images.reshape(-1, 28, 28, 1)
|
|
|
|
|
|
|
|
|
|
|
|
# 对标签进行分类编码
|
|
|
|
|
|
train_labels = to_categorical(train_labels, 10)
|
|
|
|
|
|
test_labels = to_categorical(test_labels, 10)
|
|
|
|
|
|
|
|
|
|
|
|
return train_images, train_labels, test_images, test_labels
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def lr_schedule(epoch):
|
|
|
|
|
|
lr = 1e-3
|
|
|
|
|
|
if epoch > 15:
|
|
|
|
|
|
lr *= 0.1
|
|
|
|
|
|
elif epoch > 55:
|
|
|
|
|
|
lr *= 0.01
|
|
|
|
|
|
return lr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def train_model_dp(train_images, train_labels, test_images, test_labels, dropout_rates, epochs):
|
|
|
|
|
|
lr_scheduler = LearningRateScheduler(lr_schedule)
|
|
|
|
|
|
early_stopping = EarlyStopping(monitor='val_loss',
|
|
|
|
|
|
patience=(epochs // 10 if epochs // 10 > 5 else 5),
|
|
|
|
|
|
restore_best_weights=True)
|
|
|
|
|
|
|
|
|
|
|
|
best_accuracy = 0.0
|
|
|
|
|
|
best_dropout_rate = None
|
|
|
|
|
|
best_model = None
|
|
|
|
|
|
|
|
|
|
|
|
for dropout_rate in dropout_rates:
|
|
|
|
|
|
print(f"Training model with dropout rate: {dropout_rate}")
|
|
|
|
|
|
|
|
|
|
|
|
model = LeNet5Custom(dropout_rate)
|
|
|
|
|
|
model = model.compile_model()
|
|
|
|
|
|
|
|
|
|
|
|
model.summary()
|
|
|
|
|
|
|
|
|
|
|
|
model.fit(train_images,
|
|
|
|
|
|
train_labels,
|
|
|
|
|
|
epochs=epochs,
|
|
|
|
|
|
batch_size=256,
|
|
|
|
|
|
validation_data=(test_images, test_labels),
|
|
|
|
|
|
callbacks=[lr_scheduler, early_stopping])
|
|
|
|
|
|
|
|
|
|
|
|
# 评估模型
|
|
|
|
|
|
test_loss, test_acc = model.evaluate(test_images, test_labels)
|
|
|
|
|
|
print(f'Test Accuracy with dropout rate {dropout_rate}: {test_acc:.4f} with loss {test_loss:.4f}')
|
|
|
|
|
|
|
|
|
|
|
|
if test_acc > best_accuracy:
|
|
|
|
|
|
best_accuracy = test_acc
|
|
|
|
|
|
best_dropout_rate = dropout_rate
|
|
|
|
|
|
best_model = model
|
|
|
|
|
|
|
|
|
|
|
|
return best_model, best_dropout_rate, best_accuracy
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def save_model(model, accuracy):
|
|
|
|
|
|
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S") # 获取当前时间戳并格式化为字符串
|
|
|
|
|
|
filepath = f'./model/lenet5_model_best_{timestamp}.h5'
|
|
|
|
|
|
model.save(filepath=filepath)
|
|
|
|
|
|
print(f"Best model weights saved to {filepath} with test accuracy: {accuracy:.4f}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
|
|
tf.keras.backend.clear_session()
|
|
|
|
|
|
tf.compat.v1.reset_default_graph()
|
|
|
|
|
|
|
|
|
|
|
|
train_images, train_labels, test_images, test_labels = load_and_preprocess_data()
|
|
|
|
|
|
|
|
|
|
|
|
dropout_rates = [0.1, 0.2, 0.3, 0.4, 0.5] # 0.3 is the best of [0.1, 0.2, 0.3, 0.4, 0.5]
|
|
|
|
|
|
epochs = 100
|
|
|
|
|
|
|
|
|
|
|
|
best_model, best_dropout_rate, best_accuracy = train_model_dp(train_images,
|
|
|
|
|
|
train_labels,
|
|
|
|
|
|
test_images,
|
|
|
|
|
|
test_labels,
|
|
|
|
|
|
dropout_rates,
|
|
|
|
|
|
epochs)
|
|
|
|
|
|
print(f'Best dropout rate: {best_dropout_rate} with test accuracy: {best_accuracy:.4f}')
|
|
|
|
|
|
|
|
|
|
|
|
save_model(best_model, best_accuracy)
|
