Struct GradMap 

pub struct GradMap { /* private fields */ }

Map from forward-graph node IDs to the IDs of their accumulated backward-graph gradient nodes, returned by NodeHandle::backward.

All gradient nodes live on the same graph as the forward nodes. Getting a gradient by calling GradMap::get returns a new NodeHandle handle pointing at the gradient node so callers can realize it, plug it into further computation, or back-propagate through it again.

Implementations

impl GradMap

pub fn get(&self, forward: &NodeHandle) -> Option<NodeHandle>

Get the gradient tensor for a forward tensor from the same graph.

Returns None if the forward tensor was not reachable from the root passed to backward (i.e. not part of the computation being differentiated).

Panics

Panics if any operand belongs to a different Graph than the one this builder is building into. Operand identity is graph-scoped and checked by pointer identity (Arc::ptr_eq), which cannot alias; the panic is deliberate and replaces the hang a cross-graph handle used to produce. Build every operand from one graph, or use a Result-returning entry point where one exists.

pub fn len(&self) -> usize

Number of forward nodes that have a recorded gradient.

pub fn is_empty(&self) -> bool

Whether this map has no gradients (e.g. a backward over a lone const).