python - SageMaker 中的 TensorFlow 电影镜头

Question

我是一个尝试定义 SageMaker 管道的 Tensorflow 和 Python 新手。目前，我在尝试在 SageMaker 中运行保存的模型 Movie Lens示例时遇到问题。我设法使用下面的代码训练模型，并使用 SavedModel API 将其保存到适当的 S3 存储桶中。当我加载模型并尝试使用加载的模型进行预测时，我收到错误消息：

ValueError: Could not find matching function to call loaded from the SavedModel. Got:
  Positional arguments (3 total):
    * <_VariantDataset shapes: (), types: tf.string>
    * True
    * None
  Keyword arguments: 
{}
Expected these arguments to match one of the following 0 option(s):

模型加载和预测：

    model_path = "/opt/ml/processing/model"
    tar_path = os.path.join(model_path, "model.tar.gz")
    logger.info(tar_path)
    with tarfile.open(tar_path) as tar:
        tar.extractall(path=model_path)

    logger.info("Extracted model.")
    model = tf.saved_model.load(model_path)
    scores, titles = model(tf_ratings.take(1))

模型类

class MovieLensModel(tfrs.Model):
  # We derive from a custom base class to help reduce boilerplate. Under the hood,
  # these are still plain Keras Models.

    def __init__(self,
        user_model: tf.keras.Model,
        movie_model: tf.keras.Model,
        task: tfrs.tasks.Retrieval):
        super(tfrs.Model,self).__init__() #added arguments for super

            # Set up user and movie representations.
        self.user_model = user_model
        self.movie_model = movie_model

        # Set up a retrieval task.
        self.task = task
        
    @tf.function
    def __call__(self, x, training=True, mask=None):
        user_embeddings = self.user_model(x)
        movie_embeddings = self.movie_model(X)
        
        return self.task(user_embeddings, movie_embeddings)

    def compute_loss(self, features: Dict[Text, tf.Tensor], training=False) -> tf.Tensor:
        # Define how the loss is computed.

        user_embeddings = self.user_model(features[0])
        movie_embeddings = self.movie_model(features[0])

        return self.task(user_embeddings, movie_embeddings)

模型训练并保存：

    task = tfrs.tasks.Retrieval(metrics=tfrs.metrics.FactorizedTopK(
        tf_movies.batch(128).map(movie_model)
    ))
    logger.info("Model training...")

    # Create a retrieval model.
    model = MovieLensModel(user_model, movie_model, task)
    model.compile(optimizer=tf.keras.optimizers.Adagrad(0.5))

    # Train for 3 epochs.
    model.fit(tf_ratings.batch(args.batch_size), epochs = args.epochs)

    # save mod
    model_path = os.path.join(args.model_dir, "movie_lens")
    
    logger.info("Model path is " + args.model_dir)
    model.task = tfrs.tasks.Retrieval()  # Removes the metrics.
    tf.saved_model.save(model, args.model_dir)

用于 SageMaker 容器的图像是 tensorflow-training:2.2-cpu-py37

我想上面错误中的位置参数与__call__模型类中的函数匹配。这里困扰我的是我不理解给出的错误，如果__call__在我进行预测时识别出该函数，model(value)为什么我会收到没有匹配函数的错误？

标签： pythontensorflowamazon-sagemakerserving