OneHot

Versioned name: OneHot-1

Category: Sequence processing

Short description: OneHot sets the elements in the output tensor with specified indices to on_value and fills all other locations with off_value.

Detailed description

Taking a tensor with rank N as the first input indices, OneHot produces a tensor with rank N+1 extending the original tensor with a new dimension at the axis position. The output tensor is populated with two scalar values: on_value that comes from the 3rd input and off_value that comes from the 4nd input. The population is made in the following way:

output[:, ... ,:, i, :, ... ,:] = on_value if (indices[:, ..., :, :, ..., :] == i) else off_value

where i is at the axis position in the output shape and has values from the range [0, ..., depth-1].

When some elements from the indices are greater or equal to the depth, it is a well-formed operation. The corresponding output rows are populated with off_value in this case.

The types of input scalars on_value and off_value should match and be equal to any supported type. The output tensor type is derived from the on_value or the off_value, they all have the same type.

Attributes:

• axis
  • Description: axis is a new axis position in the output shape to fill with one-hot values.
  • Range of values: an integer. Negative value means counting dimension from the end.
  • Type: int
  • Default value: None
  • Required: yes

Inputs:

• 1: indices: input tensor of type T1 with non-negative indices, behavior for negative indices is undefined. Can be 0D. Required.
• 2: depth: positive scalar (0D tensor) of type T1 that specifies the number of classes and thus the size of the one-hot dimension. Required.
• 3: on_value: scalar (0D tensor) of type T2 that fills the locations in output tensor specified in indices. Required.
• 4: off_value: scalar (0D tensor) of type T2 that fills the locations not represented in indices. Required.

Outputs:

• 1: An N+1 rank tensor of type T2, where N is a rank of the input tensor indices. A new axis of the size depth is inserted at the dimension axis.

Types

• T1: int32 or int64.
• T2: any supported data type.

Examples

<layer ... type="OneHot" ...>
    <data axis="-1"/>
    <input>
        <port id="0">    <!-- indices value: [0, 3, 1, 2] -->
            <dim>4</dim>
        </port>
        <port id="1">    <!-- depth value: 3 -->
        </port>
        <port id="2">    <!-- on_value 1 -->
        </port>
        <port id="3">    <!-- off_value 2 -->
        </port>
    </input>
    <output>
        <port id="0">    <!-- output value # [[1, 2, 2], [2, 2, 2], [2, 1, 2], [2, 2, 1]] -->
            <dim>4</dim>
            <dim>3</dim>
        </port>
    </output>
</layer>

<layer ... type="OneHot" ...>
    <data axis="1"/>
    <input>
        <port id="0">    <!-- indices value: [[0, 3, 1], [1, 2, 4]] -->
            <dim>2</dim>
            <dim>3</dim>
        </port>
        <port id="1">    <!-- depth value: 3 -->
        </port>
        <port id="2">    <!-- on_value 1 -->
        </port>
        <port id="3">    <!-- off_value 0 -->
        </port>
    </input>
    <output>
        <port id="0">    <!-- output value: [[[1, 0, 0], [0, 0, 1], [0, 0, 0]], -->
            <dim>2</dim> <!--                [[0, 0, 0], [1, 0, 0], [0, 1, 0]]] -->
            <dim>3</dim>
            <dim>3</dim>
        </port>
    </output>
</layer>