Struct Gradients 

pub struct Gradients { /* private fields */ }

A memory store mapping mathematical tensor nodes to their computed derivatives.

Implementations

impl Gradients

pub fn new() -> Self

Initializes an empty gradient tracker.

pub fn get(&self, id: &TensorId) -> Option<&Buffer>

Retrieves a reference to the computed gradient buffer for a specific Tensor.

pub fn remove(&mut self, id: &TensorId) -> Option<Buffer>

Removes and takes ownership of a gradient buffer (useful for optimizers).

pub fn insert(&mut self, id: TensorId, new_grad: Buffer) -> EtensorResult<()>

Pushes a computed derivative into the store.

The Calculus Accumulation Rule: If a gradient already exists for this TensorId (meaning the tensor was used multiple times in the forward graph), the new gradient is mathematically added to the existing one.