nn.functional.one_hot¶

lucid.nn.functional.one_hot(input_: Tensor, num_classes: int = -1, dtype: Numeric | bool | None = None) → Tensor¶

The one_hot function converts a tensor of integer indices into a one-hot encoded tensor, adding a new last dimension of size num_classes. This is commonly used in classification problems where categorical labels need to be converted to a binary format.

Function Signature¶

def one_hot(
    input_: Tensor,
    num_classes: int = -1,
    dtype: Numeric | bool | None = None
) -> Tensor

Parameters¶

input_ (Tensor): A tensor of integers representing class indices. Must contain only non-negative integer values.
num_classes (int, optional): The total number of classes. If set to -1 (default), the number of classes is inferred from the maximum index in the input (i.e., input_.max() + 1).
dtype (Numeric | bool | None, optional): The desired output data type. If None (default), the one-hot output will use the default floating type.

Returns¶

Tensor: A tensor of shape (*input_.shape, num_classes), where each index in the input is converted into a one-hot encoded vector.

One-Hot Encoding Logic¶

For each index in input_, the function sets the corresponding position in the last dimension to 1:

\[\begin{split}\text{output}[i_1, i_2, \dots, i_n, c] = \begin{cases} 1, & \text{if } c = \text{input_}[i_1, i_2, \dots, i_n] \\ 0, & \text{otherwise} \end{cases}\end{split}\]

If num_classes is not specified, it is inferred as the maximum value in input_ plus one.

Examples¶

Basic one-hot encoding:

>>> import lucid
>>> import lucid.nn.functional as F
>>> input_ = lucid.Tensor([[0, 2], [1, 3]], dtype=int)
>>> output = F.one_hot(input_, num_classes=4)
>>> print(output.shape)
(2, 2, 4)

>>> print(output)
Tensor([
    [[1, 0, 0, 0],
     [0, 0, 1, 0]],
    [[0, 1, 0, 0],
     [0, 0, 0, 1]]
])

Inferring `num_classes`:

>>> input_ = lucid.Tensor([0, 1, 2])
>>> output = F.one_hot(input_)
>>> print(output.shape)
(3, 3)

>>> print(output)
Tensor([
    [1, 0, 0],
    [0, 1, 0],
    [0, 0, 1]
])

Specifying a custom dtype:

>>> output = F.one_hot(input_, num_classes=4, dtype=lucid.Bool)
>>> print(output.dtype)
bool

Note

The input tensor must contain integers only.
The returned tensor will have one more dimension than the input: the last dimension will be num_classes.
If input_ contains a value greater than or equal to num_classes, an error will be raised.