模型保存和接续训练

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras import layers, optimizers, losses
import matplotlib.pyplot as plt
x_data = [[1, 2, 3], [2, 3, 4], [3, 4, 5]]
y_data = [1.1, 2.11, 3.09]

# 模型创建 Sequential layers.Dense
model = Sequential()
model.add(layers.Dense(units=3, input_dim=3))
model.add(layers.Dense(units=1))
model.summary()
# 模型编译 优化器，损失函数
model.compile(optimizer=optimizers.Adagrad(learning_rate=0.01), loss=losses.mean_squared_error)
# 模型训练
train_log = model.fit(x_data, y_data, epochs=200) # model.fit返回模型训练过程日志
loss_list = train_log.history['loss'] # 获取训练过程中的损失值
plt.plot(loss_list)
plt.show()
# 模型评估函数 evaluate
loss = model.evaluate(x_data,y_data)
print("模型最终损失值：", loss)

# 
# 接续训练
# 仅保存参数
model.save_weights('model_weights')
# 参数加载，接续训练
model.load_weights('model_weights')
model.fit(x_data, y_data, epochs=100)
loss = model.evaluate(x_data, y_data)
print("loss:",loss)

# 同时保存模型和参数
model.save('model.h5')
# 模型+参数加载进来并且接续训练
model_new = tf.keras.models.load_model('model.h5')
model_new.fit(x_data, y_data, epochs=100)
loss = model_new.evaluate(x_data, y_data)
print("loss:",loss)

#线性回归acc:使用mse