Struct CrossEntropy

pub struct CrossEntropy<'a> {
    pub weights: Option<&'a Array>,
    pub axis: i32,
    pub label_smoothing: f32,
    pub reduction: LossReduction,
}

Cross entropy loss function.

Fields

weights: Option<&'a Array>

Weights for each target

axis: i32

The axis over which to compute softmax. Default to CrossEntropy::DEFAULT_AXIS

label_smoothing: f32

The label smoothing factor, range [0, 1). Default to CrossEntropy::DEFAULT_LABEL_SMOOTHING

reduction: LossReduction

Reduction type. Default to CrossEntropy::DEFAULT_REDUCTION

Implementations

impl<'a> CrossEntropy<'a>

pub fn new() -> Result<Self, <<Self as Buildable>::Builder as Builder<Self>>::Error>

Creates a new instance of CrossEntropy.

impl<'a> CrossEntropy<'a>

pub const DEFAULT_AXIS: i32 = -1i32

Default value for the axis parameter.

pub const DEFAULT_LABEL_SMOOTHING: f32 = 0f32

Default value for the label_smoothing parameter.

pub const DEFAULT_REDUCTION: LossReduction = LossReduction::None

Default value for the reduction parameter.

pub const DEFAULT_WEIGHTS: Option<&'a Array> = None

Default value for the weights parameter.

pub fn apply( &self, logits: impl AsRef<Array>, targets: impl AsRef<Array>, ) -> Result<Array, Exception>

Apply the cross entropy loss function on the given logits and targets.

Params

• logits: unnormalized predicted logits
• targets: target values, as class indices

Trait Implementations

impl<'a> Buildable for CrossEntropy<'a>

type Builder = CrossEntropyBuilder<'a>

The builder type for this buildable type

impl<'a> Builder<CrossEntropy<'a>> for CrossEntropyBuilder<'a>

type Error = CrossEntropyBuildError

Error with building

fn build(self) -> Result<CrossEntropy<'a>, Self::Error>

Build the type

impl<'a> Clone for CrossEntropy<'a>

fn clone(&self) -> CrossEntropy<'a>

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Debug for CrossEntropy<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.