NameError:未定义名称“”elmo“”tensorflow集线器python

Question

我正在使用tensorflow==1.15.5。

我正在加载这样的预训练模型

# loading pre trained model to extract elmo vectors
# this is required to load custom model    
elmo = hub.Module("https://tfhub.dev/google/elmo/2", trainable=True)
class DeepLearningBot:

    dirName = os.path.dirname(__file__)
    stopwords = ['i',  'am', 'a',
                 'to', 'is', 'do', 'in', 'of']
    labelModelPath = os.path.join(dirName, "config/labels_mapping.pkl")
    modelPath = os.path.join(dirName,"models/elmo-model-bilstm2.h5")

    def init():
        Logger.log("Using tensorflow : " + tf.__version__)
        DeepLearningBot.modelBlstm = tf.keras.models.load_model(DeepLearningBot.modelPath)

要加载自定义模型，需要elmo变量(需要在模型内部使用elmo变量)

我已经在类外部初始化了变量elmo，但是当我在init()方法内部加载模型时，它会说

    return deserialize(config, custom_objects=custom_objects)
  File "c:\\env37\lib\site-packages\tensorflow_core\python\keras\layers\serialization.py", line 105, in deserialize
    printable_module_name='layer')
  File "c:\PERSONAL\projects\AdamAI\adam-chat-server\env37\lib\site-packages\tensorflow_core\python\keras\utils\generic_utils.py", line 191, in deserialize_keras_object
    list(custom_objects.items())))
  File "c:\env37\lib\site-packages\tensorflow_core\python\keras\engine\network.py", line 1081, in from_config
    process_node(layer, node_data)
  File "c:\PERSONAL\projects\AdamAI\adam-chat-server\env37\lib\site-packages\tensorflow_core\python\keras\engine\network.py", line 1039, in process_node
    layer(input_tensors, **kwargs)
  File "c:\env37\lib\site-packages\tensorflow_core\python\keras\engine\base_layer.py", line 854, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "c:\env37\lib\site-packages\tensorflow_core\python\keras\layers\core.py", line 789, in call
    return self.function(inputs, **arguments)
  File "<ipython-input-45-e850aa69f190>", line 3, in ELMoEmbedding_bl
NameError: name 'elmo' is not defined

现在我该如何解决这个问题呢？

有没有其他方法来配置预先训练好的模型？

Answer 1

模型使用的是变量elmo和tf，因此它希望在加载模型之前声明这些变量。

在使用custom_objects属性加载模型时，还有其他方法可以传递它

DeepLearningBot.modelBlstm = tf.keras.models.load_model(DeepLearningBot.modelPath,
                                                                custom_objects={
                                                                    'elmo': hub.Module("https://tfhub.dev/google/elmo/2", trainable=False),
                                                                    'tf': tf
                                                                })