Tensorflow模型使用Flask抛出此图的一个元素

Question

我有这样一个类，用于查找给定句子的句子嵌入

class Embeddings:
    def __init__(self):
        self.embedding_model_url = config_obj.tf_model_url
        self.embedding_model = hub.Module(self.embedding_model_url)
        self.messages = tf.placeholder(dtype=tf.string, shape=[None])
        self.output = self.embedding_model(self.messages)
        # self.initialize_graph()

    @staticmethod
    def initialize_graph():
        with tf.Session() as session:
            session.run([tf.global_variables_initializer(), tf.local_variables_initializer()])

    def get_sentence_embeddings(self, sentence):

        with tf.Session() as session:
            session.run([tf.global_variables_initializer(), tf.tables_initializer()])
            result = session.run(self.output, feed_dict={self.messages: [sentence]})

        return result


if __name__ == '__main__':
    sentence = "GoAir is waiving cancellation and change fees for Bhubaneswar, Kolkata and Ranchi flights for travel between May 2 and May 5, the airline said in a statement"
    tf_object = Embeddings()
    embeddings = tf_object.get_sentence_embeddings(sentence)
    print(embeddings)

这是一个独立的应用程序，但当我尝试将其与Flask集成时，如下所示

from sentence_embeddings import Embeddings

embedding_obj = Embeddings()
@app.route('/get-similar-claims', methods=['POST'])
def get_similar_claims():
    params = request.get_json()
    claim = params.get("claim", "")
    num_results = params.get("num_results", 10)
    t0 = time.time()
    # claim_embeddings = ""
    claim_embeddings = embedding_obj.get_sentence_embeddings(claim)
    logger.info("Time taken to calculate sentence embeddings - {}".format(round(time.time() - t0, 4)))
    return Response(json.dumps(claim_embeddings), mimetype='application/json')

if __name__ == '__main__':
    app.run('0.0.0.0', 5001)

它抛出一个错误

File "/Users/anuragsharma/claim_similarity/api/app.py", line 32, in get_similar_claims
    claim_embeddings = embedding_obj.get_sentence_embeddings(claim)
  File "/Users/anuragsharma/claim_similarity/api/sentence_embeddings.py", line 31, in get_sentence_embeddings
    result = session.run(self.output, feed_dict={self.messages: tf.convert_to_tensor(sentence)})
  File "/Users/anuragsharma/anaconda3/envs/similarity-search/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 929, in run
    run_metadata_ptr)
  File "/Users/anuragsharma/anaconda3/envs/similarity-search/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 1095, in _run
    'Cannot interpret feed_dict key as Tensor: ' + e.args[0])
TypeError: Cannot interpret feed_dict key as Tensor: Tensor Tensor("Placeholder:0", shape=(?,), dtype=string) is not an element of this graph.
I0730 16:44:33.384036 123145584836608 _internal.py:122] 127.0.0.1 - - [30/Jul/2019 16:44:33] "POST /get-similar-claims HTTP/1.1" 500 -

Answer 1

问题是，由于flask中的线程，您正在失去tf图的上下文。当您加载模型时，您需要保存对tf图的引用，以便以后使用它。

def __init__():
    self.model = load_model()
    self.graph = tf.get_default_graph()

def predict():
    with self.graph.as_default():
        self.model.predict()