1,保存模型:
my_model = create_model_function( ...... )
my_model.compile( ...... )
my_model.fit( ...... )
model_name . save( filepath, overwrite: bool=True, include_optimizer: bool=True )
filepath:保存的路径
overwrite:如果存在源文件,是否覆盖
include_optimizer:是否保存优化器状态
ex : mymodel.save(filepath="p402/my_model.h5", includeoptimizer=False)
2, 载入模型:
my_model = keras . models . load_model( filepath )
载入后可以继续训练:
my_model . fit( X_train_2,Y_train_2 )
也可以直接评估:
preds = my_model . evaluate( X_test, Y_test )
print ( "Loss = " + str( preds[0] ) )
print ( "Test Accuracy = " + str( preds[1] ) )
3, 如果仅保存模型的结构,而不包含其权重或配置信息,可以使用:
# save as JSON
json_string = model.to_json()
# save as YAML
yaml_string = model.to_yaml()
从保存好的json文件或yaml文件中载入模型:
# model reconstruction from JSON:
from keras.models import model_from_json
model = model_from_json(json_string)
# model reconstruction from YAML
model = model_from_yaml(yaml_string)
4,如果需要保存模型的权重,可通过下面的代码利用HDF5进行保存:
model.save_weights('my_model_weights.h5')
若在代码中初始化一个完全相同的模型,请使用:
model.load_weights('my_model_weights.h5')
5,若要加载权重到不同的网络结构(有些层一样)中,例如fine-tune或transfer-learning,可通过层名字来加载模型:
model.load_weights('my_model_weights.h5', by_name=True)
如:
"""
假如原模型为:
model = Sequential()
model.add(Dense(2, input_dim=3, name="dense_1"))
model.add(Dense(3, name="dense_2"))
...
model.save_weights(fname)
"""
# new model
model = Sequential()
model.add(Dense(2, input_dim=3, name="dense_1")) # will be loaded
model.add(Dense(10, name="new_dense")) # will not be loaded
# load weights from first model; will only affect the first layer, dense_1.
model.load_weights(fname, by_name=True)