如何將保存的 model 轉換或加載到 TensorFlow 或 Keras？

Question

我使用 tensorflow keras 創建了一個 model 並定義了一個回調來保存每個 Z20F35E630DAF439DBFA4C3F6Depoch58CZ 之后的回調。 它工作並以pb格式保存了 model 但我無法將其再次加載到 keras 因為 keras 只接受h5格式。

我有兩個問題：

• 除了 tensorflow 服務之外，我如何將保存的 model 加載到 keras/tensorflow 中？
• 如何在每個時期后以h5格式保存 keras model ？

我的回調並保存 model：

from tensorflow.keras.callbacks import ModelCheckpoint

cp_callback = ModelCheckpoint(filepath=checkpoint_path, save_freq= 'epoch', verbose=1 )

regressor.compile(optimizer = 'adam', loss = 'mean_squared_error')
regressor.fit(X_train, y_train, epochs = 10, batch_size = 32, callbacks=[cp_callback])

我保存的 model 結構：

saved_trained_10_epochs
├── assets
├── saved_model.pb
└── variables
    ├── variables.data-00000-of-00001
    └── variables.index

更新

我嘗試使用latest_checkpoint如下，但出現以下錯誤：

from tensorflow.train import latest_checkpoint

loaded_model = latest_checkpoint(checkpoint_path)
loaded_model.summary()

錯誤：

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-57-76a8ebe4f259> in <module>
----> 1 loaded_model.summary()

AttributeError: 'NoneType' object has no attribute 'summary'

在重新創建 model 之后：

loaded_regressor = Sequential()

loaded_regressor.add(LSTM(units = 180, return_sequences = True, input_shape = (X_train.shape[1], 3)))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(LSTM(units = 180, return_sequences = True))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(LSTM(units = 180, return_sequences = True))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(LSTM(units = 180, return_sequences = True))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(LSTM(units = 180, return_sequences = True))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(LSTM(units = 180))
loaded_regressor.add(Dropout(0.2))

loaded_regressor.add(Dense(units = 1))

loaded_regressor.compile(optimizer = 'adam', loss = 'mean_squared_error')
loaded_regressor.load_weights(latest_checkpoint(checkpoint_path))

錯誤：

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-30-c344f1759d01> in <module>
     22 
     23 loaded_regressor.compile(optimizer = 'adam', loss = 'mean_squared_error')
---> 24 loaded_regressor.load_weights(latest_checkpoint(checkpoint_path))

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in load_weights(self, filepath, by_name)
    160         raise ValueError('Load weights is not yet supported with TPUStrategy '
    161                          'with steps_per_run greater than 1.')
--> 162     return super(Model, self).load_weights(filepath, by_name)
    163 
    164   @trackable.no_automatic_dependency_tracking

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/network.py in load_weights(self, filepath, by_name)
   1375             format.
   1376     """
-> 1377     if _is_hdf5_filepath(filepath):
   1378       save_format = 'h5'
   1379     else:

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/network.py in _is_hdf5_filepath(filepath)
   1670 
   1671 def _is_hdf5_filepath(filepath):
-> 1672   return (filepath.endswith('.h5') or filepath.endswith('.keras') or
   1673           filepath.endswith('.hdf5'))
   1674 

AttributeError: 'NoneType' object has no attribute 'endswith'

Answer 1

tf.keras models are loaded using tf.keras.models.load_model , this should work fine as tf.keras supports reading/writing multiple formats, including tensorflow checkpoints.