Tensors management

Machine Learning Web API in Tizen requires TensorsInfo and TensorsData objects. These objects are used to perform calculation with neural network models.

The main features of the Machine Learning API include the following:

• Checking support of a specific neural network framework

  You can check NNFW availability using checkNNFWAvailability.

• Creating tensors structure

  You can use the TensorsInfo object to manage the tensor information.

• Reading and writing raw data

  You can read and write tensor data using the TensorsData object.

Check NNFW availability

To check whether specific Neural Network Framework (NNFW) is supported, you can use checkNNFWAvailability:

var fw = "TENSORFLOW_LITE";
var hw = "CPU";
var available = tizen.ml.checkNNFWAvailability(fw, hw);
console.log(available); // true

Manage tensor information

1. To configure the tensor information, you need to create a new instance of the TensorsInfo class:

   var tensorsInfo = new tizen.ml.TensorsInfo();
   

2. TensorsInfo object can store information such as name, data type, and dimensions:

   tensorsInfo.addTensorInfo("tensor", "UINT8", [2, 2]);
   

   Note

   A single TensorsInfo object can store information about up to 16 tensors.

3. You can modify tensor’s parameters with help of setTensorName, setTensorType, and setDimensions:

   tensorsInfo.setTensorName(0, "tensorName");
   tensorsInfo.setTensorType(0, "UINT16");
   tensorsInfo.setTensorDimensions(0, [2, 2]);
   

4. To clone and compare TensorsInfo object:

   var clone = tensorsInfo.clone();
   var isEqual = tensorsInfo.equals(clone);
   

   Note

   Use equals() to compare TensorInfo objects. Comparisons must not be done with built-in ==, ===, !=, !== operators which may return false results for this data type.

5. To calculate the byte size of tensor data use getTensorSize:

   var byteSize = tensorsInfo.getTensorSize(0);
   

6. After you create tensor definition, you can use getTensorsData to get TensorsData object:

   var tensorsData = tensorsInfo.getTensorsData();
   

7. Ensure to dispose off the TensorsInfo object when you do not need it anymore:

   tensorsInfo.dispose();
   

Manage tensor data

The TensorsData object keeps data value of the tensors.

1. To get specific data object from TensorsData, use getTensorRawData:

   var rawData = tensorsData.getTensorRawData(0);
   

   • You can also specify the location and size of data you want to get:

     // fetch one element
     var rawDataOne = tensorsData.getTensorRawData(0, [0, 0], [1, 1]);
     // fetch first row
     var rawDataRow = tensorsData.getTensorRawData(0, [0, 0], [-1, 1]);
     

   Note

   To gather more information about specifying location and size of the raw data, see TensorsData.getTensorRawData (in TV applications).

2. To set data to TensorsData object, use setTensorRawData:

   tensorsData.setTensorRawData(0, [1, 2, 3, 4]);
   

   • You can also specify the location and size of data you want to set:

     // set only one element
     tensorsData.setTensorRawData(0, [7], [0, 0], [1, 1]);
     // set first row
     tensorsData.setTensorRawData(0, [4, 5], [0, 0], [-1, 1]);
     

   Note

   To gather more information about specifying location and size of the raw data, see TensorsData.setTensorRawData (in TV applications).

3. Ensure to dispose off the TensorsData objects when you do not need them anymore:

   tensorsData.dispose();
   

Related information

• Dependencies
  • Tizen 6.5 and Higher for TV

• API References
  • TV