Struct InferResult 

pub struct InferResult<L: LinAlg = CpuLinAlg> {
    pub y_conv: L::Vector,
    pub latent_concat: L::Vector,
    pub hidden_states: Vec<L::Vector>,
    pub prediction_errors: Vec<L::Vector>,
    pub surprise_score: f64,
    pub steps_used: usize,
    pub converged: bool,
    pub tanh_components: Vec<Option<L::Vector>>,
}

Result of the predictive coding inference loop.

Contains converged output logits, hidden state representations, and diagnostic information about the inference process.

Generic over a LinAlg backend L. Defaults to CpuLinAlg.

Fields

y_conv: L::Vector

Converged output logits.

latent_concat: L::Vector

All hidden states concatenated (fed to critic).

hidden_states: Vec<L::Vector>

Per-layer hidden state activations.

prediction_errors: Vec<L::Vector>

Per-layer prediction errors from the last PC inference step. Ordered from top hidden layer to bottom (reverse layer order).

surprise_score: f64

RMS prediction error across layers.

steps_used: usize

Number of inference steps performed.

converged: bool

Whether the inference loop converged within tolerance.

tanh_components: Vec<Option<L::Vector>>

Per-layer tanh components for residual layers. None for non-skip layers, Some(tanh_out) for skip-eligible layers. Needed for correct backward pass (derivative on tanh_out, not full h[i]).

Trait Implementations

impl<L: Clone + LinAlg> Clone for InferResult<L>

where L::Vector: Clone,

fn clone(&self) -> InferResult<L>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<L: Debug + LinAlg> Debug for InferResult<L>

where L::Vector: Debug,

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

Formats the value using the given formatter.