将 TensorFlow JS model.predict 的值转化为变量

Question

I am trying to implement a neural network in JS, which can predict certain values and then store these values in a variable for later use.

async function processModel(inputs) {
  const model = await tf.loadLayersModel(modelURL);
  inputs = tf.tensor(inputs);

  var predictions = model.predict(inputs);
  predictions = predictions.dataSync();

  console.log(typeof(predictions));

  return predictions;
}

Launching this informs me that predictions is of type object . I would like to get the values of the Promise which is returned and store these values in a variable, so I can compare them to other variables.

Can anyone help me with this please?

Answer 1

prredictions in an array. The array indexing is the way to get the predicted value

Answer 2

According to datasync method , variable.dataSync() returns DataTypeMap[NumericDataType] . It returns a TypedArray of any NumericDataType .

According to MDN documentation on TypedArray ,

a TypedArray object describes an array-like view of an underlying binary data buffer.

So, you are correct that console.log(typeof(predictions)) will print object in the console.

But as predictions is a TypedArray , you can use it like regular JavaScript array. You can print predictions[0] and so on.

Let's see an example of TypedArray :

 // create a TypedArray with a size in bytes const typedArrayExample = new Float32Array(2); typedArrayExample[0] = 32.36; console.log(typeof(typedArrayExample)); // object console.log(typedArrayExample); // {"0": 32.36000061035156, "1": 0} console.log(typedArrayExample[0]); // 32.36000061035156

Reference:

• Documentation on datasync method
• MDN documentation on TypedArray
• Documentation on predict method

Answer 3

You can access the predicted values in the output Tensor using the arraySync() method. For eg if the output Tensor is a two dimensional array with only a single output value, then the code below will give you the predicted value.

 const resultTensor = model.predict(tf.tensor2d([50], [1, 1]));
 const predictedValue = resultTensor.arraySync()[0][0];
 console.log(predictedValue);