DML_ONE_HOT_OPERATOR_DESC structure (directml.h)

Produces a tensor filled with one-hot encoded values. This operator produces an output tensor where, for all sequences in a chosen axis, all but one element in that sequence is set to OffValue, and the remaining single element is set to OnValue. A sequence refers to one of the sets of elements that exist along the Axis dimension of the OutputTensor.

The location of the OnValue for each sequence and the choice of OnValue/OffValue are determined by the IndicesTensor and ValuesTensor, respectively.

Syntax

struct DML_ONE_HOT_OPERATOR_DESC {
  const DML_TENSOR_DESC *IndicesTensor;
  const DML_TENSOR_DESC *ValuesTensor;
  const DML_TENSOR_DESC *OutputTensor;
  UINT                  Axis;
};

Members

IndicesTensor

Type: const DML_TENSOR_DESC*

A tensor containing the index in elements of the OnValue, for each sequence along the Axis. Indices are measured relative to the beginning of their sequence (as opposed to the beginning of the tensor). For example, an index of 0 always refers to the first element for all sequences in an axis.

If an index value for a sequence exceeds the number of elements along the Axis dimension in the OutputTensor, then that index value is ignored, and all elements in that sequence will be set to OffValue.

Starting with DML_FEATURE_LEVEL_3_0, this operator supports negative index values when using a signed integral type with this tensor. Negative indices are interpreted as being relative to the end of the sequence. For example, an index of -1 refers to the last element in the sequence.

This tensor must have dimension count and sizes equal to the OutputTensor, except for the dimension specified by the Axis parameter. The size of the Axis dimension must be 1. For example if the OutputTensor has sizes of {2,3,4,5} and Axis is 1, the sizes of the IndicesTensor must be {2,1,4,5}.

ValuesTensor

Type: const DML_TENSOR_DESC*

This tensor may have any size, so long as it contains at least two elements. The 0th element of this tensor is interpreted as the OffValue, and the 1st element along the fastest-changing dimension of size >1 is interpreted as the OnValue.

OutputTensor

Type: const DML_TENSOR_DESC*

The output tensor to write the results to. This tensor must have dimension count and sizes equal to the IndicesTensor, except for the dimension specified by the Axis parameter. The size of the Axis dimension in this tensor may have any value greater than 0.

Axis

Type: UINT

The index of the dimension to produce one-hot encoded sequences along. This value must be less than the DimensionCount of the IndicesTensor.

Examples

Example 1

IndicesTensor: (Sizes:{1,1,3,1}, DataType:UINT32)
[[[[0],
   [3],
   [2]]]]
   
ValuesTensor: (Sizes:{1,1,1,2}, DataType:FLOAT32)
[[[[0, 1]]]]

Axis: 3

OutputTensor: (Sizes:{1,1,3,4}, DataType:FLOAT32)
[[[[1, 0, 0, 0],    // The one-hot encoding is formed across the rows
   [0, 0, 0, 1],
   [0, 0, 1, 0]]]]

Example 2. Using a different axis

IndicesTensor: (Sizes:{1,1,1,4}, DataType:UINT32)
[[[[0, 2, 1, 0]]]]
   
ValuesTensor: (Sizes:{1,1,1,2}, DataType:FLOAT32)
[[[[0, 1]]]]

Axis: 2

OutputTensor: (Sizes:{1,1,3,4}, DataType:FLOAT32)
[[[[1, 0, 0, 1],    // The one-hot encoding is formed across the columns
   [0, 0, 1, 0],
   [0, 1, 0, 0]]]]

Example 3. Using different on/off values

IndicesTensor: (Sizes:{1,1,3,1}, DataType:UINT32)
[[[[0],
   [3],
   [2]]]]
   
ValuesTensor: (Sizes:{1,1,3,1}, DataType:FLOAT32)
[[[[4],    // off value
   [2],    // on value
   [9]]]] // unused

Axis: 3

OutputTensor: (Sizes:{1,1,3,4}, DataType:FLOAT32)
[[[[2, 4, 4, 4],
   [4, 4, 4, 2],
   [4, 4, 2, 4]]]]

Example 4. Negative and out-of-bounds indices

IndicesTensor: (Sizes:{1,1,3,1}, DataType:INT32)
[[[[ -3],
   [100],
   [  3]]]]
   
ValuesTensor: (Sizes:{1,1,1,2}, DataType:FLOAT32)
[[[[0, 1]]]]

Axis: 3

OutputTensor: (Sizes:{1,1,3,4}, DataType:FLOAT32)
[[[[0, 1, 0, 0],    // negative indices count from the end
   [0, 0, 0, 0],    // out-of-bounds indices are ignored; all elements are set to OffValue
   [0, 0, 0, 1]]]]

Remarks

Availability

This operator was introduced in DML_FEATURE_LEVEL_2_0.

Tensor constraints

• IndicesTensor, OutputTensor, and ValuesTensor must have the same DimensionCount.
• OutputTensor and ValuesTensor must have the same DataType.

Tensor support

DML_FEATURE_LEVEL_4_1 and above

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_2_1 and above

DML_FEATURE_LEVEL_2_0 and above

Requirements