Struct Gradients 

pub struct Gradients { /* private fields */ }

Move-only accumulated gradient bundle backed by one allocation group.

Examples

use tenferro_ad::{EagerRuntime, EagerTensor, Tensor};
use tenferro_cpu::CpuBackend;

let ctx = EagerRuntime::with_cpu_backend(CpuBackend::new())?;
let x = EagerTensor::requires_grad_in(
    Tensor::from_vec_col_major(vec![2], vec![2.0_f64, 3.0])?,
    ctx,
)?;
let loss = x.mul(&x)?.reduce_sum(Some(&[0]))?;
let gradients = loss.backward()?;
assert!(!gradients.is_empty());

Implementations

impl Gradients

pub fn len(&self) -> usize

Return the number of retained gradient descriptors.

pub fn is_empty(&self) -> bool

Return whether no gradient was produced.

pub fn grad(&self, key: &ValueKey<StdTensorOp>) -> Option<TensorView<'_>>

Borrow one gradient view by its local value key.

pub fn take_grad( &mut self, key: &ValueKey<StdTensorOp>, ) -> Result<Option<Tensor>>

Consume one gradient owner while leaving the bundle unchanged on failure.

Errors

Returns tenferro_tensor::Error::RuntimeState when the descriptor is invalid or its allocation is aliased. A missing key is reported as Ok(None).

Trait Implementations

impl Debug for Gradients

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

Formats the value using the given formatter.