深度學習TensorFlow2：如何使用keras模型保存和序列化？

1.保存序列模型和函數模型

# 構建一個簡單的模型並訓練
from __future__ import absolute_import, division, print_function
import tensorflow as tf
tf.keras.backend.clear_session()
from tensorflow import keras
from tensorflow.keras import layers
​
inputs = keras.Input(shape=(784,), name='digits')
x = layers.Dense(64, activation='relu', name='dense_1')(inputs)
x = layers.Dense(64, activation='relu', name='dense_2')(x)
outputs = layers.Dense(10, activation='softmax', name='predictions')(x)
​
model = keras.Model(inputs=inputs, outputs=outputs, name='3_layer_mlp')
model.summary()
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(60000, 784).astype('float32') / 255
x_test = x_test.reshape(10000, 784).astype('float32') / 255
​
model.compile(loss='sparse_categorical_crossentropy',
              optimizer=keras.optimizers.RMSprop())
history = model.fit(x_train, y_train,
                    batch_size=64,
                    epochs=1)
​
predictions = model.predict(x_test)
Model: "3_layer_mlp"
_________________________________________________________________
Layer (type)                 Output Shape              Param # 
=================================================================
digits (InputLayer)          [(None, 784)]             0         
_________________________________________________________________
dense_1 (Dense)              (None, 64)                50240     
_________________________________________________________________
dense_2 (Dense)              (None, 64)                4160      
_________________________________________________________________
predictions (Dense)          (None, 10)                650       
=================================================================
Total params: 55,050
Trainable params: 55,050
Non-trainable params: 0
_________________________________________________________________
60000/60000 [==============================] - 2s 29us/sample - loss: 0.3116

1.1保存全模型

可以對整個模型進行保存，其保存的內容包括：

• 該模型的架構
• 模型的權重（在訓練期間學到的）
• 模型的訓練配置（你傳遞給編譯的），如果有的話
• 優化器及其狀態（如果有的話）（這使您可以從中斷的地方重新啟動訓練）

 

import numpy as np
model.save('the_save_model.h5')
new_model = keras.models.load_model('the_save_model.h5')
new_prediction = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_prediction, atol=1e-6) # 預測結果一樣
​

1.2 保存為SavedModel文件

keras.experimental.export_saved_model(model, 'saved_model')
new_model = keras.experimental.load_from_saved_model('saved_model')
new_prediction = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_prediction, atol=1e-6) # 預測結果一樣
​

 

1.3僅保存網絡結構

僅保持網絡結構，這樣導出的模型並未包含訓練好的參數

config = model.get_config()
reinitialized_model = keras.Model.from_config(config)
new_prediction = reinitialized_model.predict(x_test)
assert abs(np.sum(predictions-new_prediction)) >0

也可以使用json保存網絡結構

json_config = model.to_json()
reinitialized_model = keras.models.model_from_json(json_config)
new_prediction = reinitialized_model.predict(x_test)
assert abs(np.sum(predictions-new_prediction)) >0

1.4僅保存網絡參數

weights = model.get_weights()
model.set_weights(weights)
# 可以把結構和參數保存結合起來
config = model.get_config()
weights = model.get_weights()
new_model = keras.Model.from_config(config) # config只能用keras.Model的這個api
new_model.set_weights(weights)
new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)

1.5完整的模型保存方法

json_config = model.to_json()
with open('model_config.json', 'w') as json_file:
    json_file.write(json_config)
​
model.save_weights('path_to_my_weights.h5')
​
with open('model_config.json') as json_file:
    json_config = json_file.read()
new_model = keras.models.model_from_json(json_config)
new_model.load_weights('path_to_my_weights.h5')
​
new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)
# 當然也可以一步到位
model.save('path_to_my_model.h5')
del model
model = keras.models.load_model('path_to_my_model.h5')

1.6保存網絡權重為SavedModel格式

model.save_weights('weight_tf_savedmodel')
model.save_weights('weight_tf_savedmodel_h5', save_format='h5')

1.7子類模型參數保存

子類模型的結構無法保存和序列化，只能保持參數

# 構建模型
class ThreeLayerMLP(keras.Model):
  
    def __init__(self, name=None):
        super(ThreeLayerMLP, self).__init__(name=name)
        self.dense_1 = layers.Dense(64, activation='relu', name='dense_1')
        self.dense_2 = layers.Dense(64, activation='relu', name='dense_2')
        self.pred_layer = layers.Dense(10, activation='softmax', name='predictions')
​
    def call(self, inputs):
        x = self.dense_1(inputs)
        x = self.dense_2(x)
        return self.pred_layer(x)
​
def get_model():
    return ThreeLayerMLP(name='3_layer_mlp')
​
model = get_model()
# 訓練模型
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(60000, 784).astype('float32') / 255
x_test = x_test.reshape(10000, 784).astype('float32') / 255
​
model.compile(loss='sparse_categorical_crossentropy',
              optimizer=keras.optimizers.RMSprop())
history = model.fit(x_train, y_train,
                    batch_size=64,
                    epochs=1)
60000/60000 [==============================] - 2s 28us/sample - loss: 0.3217

 

# 保存權重參數
model.save_weights('my_model_weights', save_format='tf')
​
# 輸出結果，供后面對比
​
predictions = model.predict(x_test)
first_batch_loss = model.train_on_batch(x_train[:64], y_train[:64])
​
# 讀取保存的模型參數
new_model = get_model()
new_model.compile(loss='sparse_categorical_crossentropy',
                  optimizer=keras.optimizers.RMSprop())
​
#new_model.train_on_batch(x_train[:1], y_train[:1])
​
new_model.load_weights('my_model_weights')
​
new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)
​
​
new_first_batch_loss = new_model.train_on_batch(x_train[:64], y_train[:64])
assert first_batch_loss