Struct easy_ml::differentiation::Derivatives

pub struct Derivatives<T> { /* private fields */ }

Computed derivatives of a computational graph for some output Record variable.

This can be indexed by any Record used in the computational graph to get the derivative with respect to that input.

Indexing using Records not involved in the computational graph, or involved in a different one will return nonsense or index out of bounds and panic. In the future this may be changed to always panic.

Implementations

impl<T: Clone + Primitive> Derivatives<T>

pub fn at_tensor_index<S, const D: usize>( &self, indexes: [usize; D], input: &RecordTensor<'_, T, S, D> ) -> Option<T>

where S: TensorRef<(T, Index), D>,

Queries the derivative at the provided index into the record tensor as input.

If you construct a Derivatives object for some output y, and call .at_tensor_index(i, &xs) on it for some input container xs and index i, this returns dy/dx where x = xs[i].

If the index into the tensor is invalid, returns None instead.

pub fn at_tensor<S, const D: usize>( &self, input: &RecordTensor<'_, T, S, D> ) -> Tensor<T, D>

where S: TensorRef<(T, Index), D>,

Queries the derivatives at every element in the record tensor input.

If you construct a Derivatives object for some output y, and call .at_tensor(&xs) on it for some input container xs this returns dy/dx for every x in xs.

pub fn at_matrix_index<S>( &self, row: Row, column: Column, input: &RecordMatrix<'_, T, S> ) -> Option<T>

where S: MatrixRef<(T, Index)> + NoInteriorMutability,

Queries the derivative at the provided index into the record matrix as input.

If you construct a Derivatives object for some output y, and call .at_matrix_index(i, j, &xs) on it for some input container xs and indexes i and j, this returns dy/dx where x = xs[i, j].

If the index into the tensor is invalid, returns None instead.

pub fn at_matrix<S>(&self, input: &RecordMatrix<'_, T, S>) -> Matrix<T>

where S: MatrixRef<(T, Index)> + NoInteriorMutability,

Queries the derivatives at every element in the record matrix input.

If you construct a Derivatives object for some output y, and call .at_matrix(&xs) on it for some input container xs this returns dy/dx for every x in xs.

impl<T: Clone + Primitive> Derivatives<T>

pub fn at(&self, input: &Record<'_, T>) -> T

Quries the derivative at the provided record as input.

If you construct a Derivatives object for some output y, and call .at(&x) on it for some input x, this returns dy/dx.

Trait Implementations

impl<T: Clone> Clone for Derivatives<T>

Any derivatives of a Cloneable type implements clone

fn clone(&self) -> Self

Returns a copy of the value.

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

Performs copy-assignment from source.

impl<T: Debug> Debug for Derivatives<T>

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

Formats the value using the given formatter.

impl<T> From<Derivatives<T>> for Vec<T>

fn from(derivatives: Derivatives<T>) -> Self

Converts the Derivatives struct into a Vec of derivatives that can be indexed with usizes. The indexes correspond to the index field on Records.

impl<'a, T: Primitive> Index<&Record<'a, T>> for Derivatives<T>

fn index(&self, input: &Record<'a, T>) -> &Self::Output

Quries the derivative at the provided record as input.

If you construct a Derivatives object for some output y, and call .at(&x) on it for some input x, this returns dy/dx.

type Output = T

The returned type after indexing.