Struct Variable 

pub struct Variable<T: Float> { /* private fields */ }

A variable in the computation graph that wraps a Tensor and supports reverse-mode automatic differentiation.

Variable<T> uses shared ownership (Rc<RefCell<...>>) so that the same node can appear as a parent in multiple downstream operations. Cloning a Variable is cheap — it just increments the reference count.

Implementations

impl<T: Float> Variable<T>

pub fn new(data: Tensor<T>, requires_grad: bool) -> Self

Create a new leaf variable.

Examples

let t = Tensor::<f64>::from_vec(vec![1.0, 2.0, 3.0], vec![3]).unwrap();
let v = Variable::new(t, true);
assert!(v.requires_grad());
assert_eq!(v.shape(), vec![3]);

pub fn data(&self) -> Tensor<T>

Return a clone of the underlying tensor data.

pub fn shape(&self) -> Vec<usize>

Return the shape of the underlying tensor.

pub fn grad(&self) -> Option<Tensor<T>>

Return the accumulated gradient, if any.

pub fn requires_grad(&self) -> bool

Whether this variable tracks gradients.

pub fn zero_grad(&self)

Reset the gradient to None.

pub fn detach(&self) -> Self

Detach from the computation graph, returning a new leaf variable with the same data but no graph history.

pub fn set_data(&self, data: Tensor<T>)

Replace the data tensor (used by optimizers and weight loading).

pub fn set_grad(&self, grad: Tensor<T>)

Replace the gradient tensor (used by GradScaler for unscaling).

pub fn backward(&self)

Run reverse-mode automatic differentiation starting from this variable.

This variable is expected to be a scalar (single-element tensor). Its gradient is seeded with 1.0. After backward(), each ancestor with requires_grad == true will have its .grad() populated.

Trait Implementations

impl<T: Float> Add for &Variable<T>

type Output = Variable<T>

The resulting type after applying the + operator.

fn add(self, rhs: Self) -> Variable<T>

Performs the + operation.

impl<T: Float> Add for Variable<T>

type Output = Variable<T>

The resulting type after applying the + operator.

fn add(self, rhs: Self) -> Variable<T>

Performs the + operation.

impl<T: Float> Clone for Variable<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Float> Debug for Variable<T>

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

Formats the value using the given formatter.

impl<T: Float> Mul for &Variable<T>

type Output = Variable<T>

The resulting type after applying the * operator.

fn mul(self, rhs: Self) -> Variable<T>

Performs the * operation.

impl<T: Float> Mul for Variable<T>

type Output = Variable<T>

The resulting type after applying the * operator.

fn mul(self, rhs: Self) -> Variable<T>

Performs the * operation.

impl<T: Float> Neg for &Variable<T>

type Output = Variable<T>

The resulting type after applying the - operator.

fn neg(self) -> Variable<T>

Performs the unary - operation.

impl<T: Float> Neg for Variable<T>

type Output = Variable<T>

The resulting type after applying the - operator.

fn neg(self) -> Variable<T>

Performs the unary - operation.

impl<T: Float> Sub for &Variable<T>

type Output = Variable<T>

The resulting type after applying the - operator.

fn sub(self, rhs: Self) -> Variable<T>

Performs the - operation.

impl<T: Float> Sub for Variable<T>

type Output = Variable<T>

The resulting type after applying the - operator.

fn sub(self, rhs: Self) -> Variable<T>

Performs the - operation.