如何解析张量流模型的字符串输出

Question

使用此处的代码创建了一个模型：https://gist.github.com/gaganmalhotra/1424bd3d0617e784976b29d5846b16b1

要获得 java 中概率的预测，可以使用以下代码完成：

  public static void main(String[] args) {
    Session session = SavedModelBundle.load("/Users/gagandeep.malhotra/Documents/SampleTF_projects/tf_iris_model/1510707746/", "serve").session();

    Tensor x =
        Tensor.create(
            new long[] {2, 4},
            FloatBuffer.wrap(
                new float[] {
                  6.4f, 3.2f, 4.5f, 1.5f,
                  5.8f, 3.1f, 5.0f, 1.7f
                }));

    final String xName = "Placeholder:0";
    final String scoresName = "dnn/head/predictions/probabilities:0";

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

    // Outer dimension is batch size; inner dimension is number of classes
    float[][] scores = new float[2][3];

    outputs.get(0).copyTo(scores);
    System.out.println(Arrays.deepToString(scores));
  }

但是如果我们想复制下面代码的预测类（字符串类型）：

final String xName = "Placeholder:0";
final String className = "dnn/head/predictions/str_classes:0";

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

// Outer dimension is batch size; inner dimension is number of classes
String[][] classes = new String[2][1];

outputs.get(0).copyTo(classes);
System.out.println(Arrays.deepToString(classes));

我最终遇到这样的错误：

Exception in thread "main" java.lang.IllegalArgumentException: cannot copy Tensor with 2 dimensions into an object with 1
    at org.tensorflow.Tensor.throwExceptionIfTypeIsIncompatible(Tensor.java:739)
    at org.tensorflow.Tensor.copyTo(Tensor.java:450)
    at deeplearning.IrisTFLoad.main(IrisTFLoad.java:71)

但维度与输出张量相同：[STRING 张量，形状为 [2, 1]]

PS: 签名定义如下 -

The given SavedModel SignatureDef contains the following input(s):
    inputs['x'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 4)
        name: Placeholder:0
    The given SavedModel SignatureDef contains the following output(s):
    outputs['class_ids'] tensor_info:
        dtype: DT_INT64
        shape: (-1, 1)
        name: dnn/head/predictions/ExpandDims:0
    outputs['classes'] tensor_info:
        dtype: DT_STRING
        shape: (-1, 1)
        name: dnn/head/predictions/str_classes:0
    outputs['logits'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 3)
        name: dnn/head/logits:0
    outputs['probabilities'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 3)
        name: dnn/head/predictions/probabilities:0
    Method name is: tensorflow/serving/predict

尝试过的事情：

张量张量=(张量)outputs.get(0); byte[][][] 结果 = tensor.copyTo(new byte[2][1][]);

但错误如下：

Exception in thread "main" java.lang.IllegalStateException: invalid DataType(7)
    at org.tensorflow.Tensor.readNDArray(Native Method)
    at org.tensorflow.Tensor.copyTo(Tensor.java:451)
    at deeplearning.IrisTFLoad.main(IrisTFLoad.java:74)

Answer 1

DT_STRING 类型的 TensorFlow 张量包含 arbitrary byte sequences 作为元素，而不是 Java Strings（字符序列）。

因此，你想要的是这样的：

byte[][][] classes = new byte[2][1][];
outputs.get(0).copyTo(classes);

如果你想获得 Java String 对象，那么你需要知道你的模型产生类的编码，然后可以做类似的事情（假设 UTF-8 编码）：

String[][] classesStrings = new String[2][1];
for (int i = 0; i < classes.length; ++i) {
  for (int j = 0; j < classes[i].length; ++j) {
    classesString[i][j] = new String(classes[i][j], UTF_8);
  }
}

希望对您有所帮助。您可能还会发现 unittest 很有启发性。

如何解析张量流模型的字符串输出

How to parse String output of a tensorflow model

python

java

tensorflow

tensorflow-serving