如何將訓練好的網路進行儲存以便以後使用, 進行後續的研究呢?
首先,定義一個簡單的LSTM模型:

from keras.models import Sequential
from keras.layers import LSTM, Dense
model = Sequential()
model.add(LSTM(4,input_shape=(1,8)))
model.add(Dense(1))

整體儲存模型及引數

首先,安裝python的h5py包.
sudo pip3 install h5py
之後呼叫model.save(filepath)將Keras模型和權重儲存在一個HDF5檔案中，該檔案將包含：
- 模型的結構，以便重構該模型
- 模型的權重
- 訓練配置（損失函式，優化器等）
- 優化器的狀態，以便於從上次訓練中斷的地方開始
使用keras.models.load_model(filepath)

from keras.models import load_model
model.save('my_model.h5')  # creates a HDF5 file 'my_model.h5'
del model  # deletes the existing model
# returns a compiled model identical to the previous one
model = load_model('my_model.h5')

只儲存模型的結構

可以用model.to_jason()將模型序列化儲存為json檔案.

# save as JSON
json_string = model.to_json()

例如上面LSTM網路的json_string就是:

json_string
Out[10]: '{"class_name": "Sequential", "config": [{"class_name": "LSTM", "config": {"name": "lstm_1", "trainable": true, "batch_input_shape": [null, 1, 8], "dtype": "float32", "return_sequences": false, "return_state"
 
: false, "go_backwards": false, "stateful": false, "unroll": false, "implementation": 0, "units": 4, "activation": "tanh", "recurrent_activation": "hard_sigmoid", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "recurrent_initializer": {"class_name": "Orthogonal", "config": {"gain": 1.0, "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "unit_forget_bias": true, "kernel_regularizer": null, "recurrent_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "recurrent_constraint": null, "bias_constraint": null, "dropout": 0.0, "recurrent_dropout": 0.0}}, {"class_name": "Dense", "config": {"name": "dense_1", "trainable": true, "units": 1, "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}], "keras_version": "2.0.8", "backend": "tensorflow"}'

裡面記錄了網路的整體結構, 各個層的引數設定等資訊. 將json字串儲存到檔案.

open('my_model_architecture.json','w').write(json_string)

當然，你也可以從儲存好的json檔案或yaml檔案中載入模型：

# 讀取json檔案
from keras.models import model_from_json
json_string = open('my_model_architecture.json').read()
model = model_from_json(json_string)

除了json格式,還可以儲存為yaml格式的字串:

# save as YAML
yaml_string = model.to_yaml()
# 類似地,讀取yaml檔案
from keras.models import model_from_yaml
model = model_from_yaml(yaml_string)

儲存模型權重等配置資訊

經過調參後網路的輸出精度比較滿意後,可以將訓練好的網路權重引數儲存下來.
可通過下面的程式碼利用HDF5進行儲存
model.save_weights(‘my_model_weights.h5’)

以後用的時候可以像這樣載入模型:
model.load_weights(‘my_model_weights.h5’)

如果你需要載入權重到不同的網路結構（有些層一樣）中，例如fine-tune或transfer-learning，你可以通過層名字來載入模型：
model.load_weights('my_model_weights.h5', by_name=True)
首先在建模時,最好對每一層都指定名字, 例如：

# 定義模型
model = Sequential()
model.add(LSTM(4, input_shape=(1, 8), name="lstm_old"))
model.add(Dense(1, name="dense_old"))
...
model.save_weights('my_model_weights.h5')

# 新模型, 過載了前一個模型訓練好的LSTM層

model_new = Sequential()
model_new.add(LSTM(4, input_shape=(1, 8), name="lstm_old"))  # will be loaded
model_new.add(Dense(10, name="dense_new"))  # will not be loaded

# 載入LSTM層訓練好的引數
model.load_weights('my_model_weights.h5', by_name=True)