Question

我按照下面提供的说明在Java中加载训练有素的Tensorflow模型：

How to import an saved Tensorflow model train using tf.estimator and predict on input data

我成功保存了我的python模型并将其导入我的java代码中。

以下是我的python代码中保存模型的部分：

# Export trained model
def serving_input_receiver_fn():
  inputs = {"my_feature": tf.placeholder(shape=[None, 1], dtype=tf.float32)}
  return tf.estimator.export.ServingInputReceiver(inputs, inputs)

export_dir = kmeans.export_savedmodel(
    export_dir_base="/tmp/path_to_model_directory",
    serving_input_receiver_fn=serving_input_receiver_fn,
    assets_extra={"file_name":"/tmp/path_to_file"
                 })

这是用于读取模型，进行预测和读取结果的java代码：

Tensor x = Tensor.create(
            new long[] {1, 1},
            FloatBuffer.wrap(
                new float[] {
                  10000.0f
                }));

final String xName = "inputs_tensor_info_name";
final String yName = "outputs_tensor_info_name";


List<Tensor<?>> outputTensor = session.runner()
        .feed(xName, x)
        .fetch(yName)
        .run();

for (int i=0; i < outputTensor.size(); i++ ) {  
    Integer[] copy = new Integer[1];
    System.out.println(outputTensor.get(i));
    System.out.println(outputTensor.get(i).copyTo(copy));
}

但是，当尝试在数组中加载预测时，会发生以下异常：

Exception in thread "main" java.lang.IllegalArgumentException: cannot create non-scalar Tensors from arrays of boxed values
at org.tensorflow.Tensor.objectCompatWithType(Tensor.java:722)
at org.tensorflow.Tensor.throwExceptionIfTypeIsIncompatible(Tensor.java:742)
at org.tensorflow.Tensor.copyTo(Tensor.java:450)
at com.pivot.maven.quickstart.HelloTF.main(HelloTF.java:105)

如果没有弄明白如何解决这个问题，如果有人指出我正确的方向，我将不胜感激。谢谢。

Tensorflow：无法从盒装值数组创建非标量张量

0 个答案: