Struct NeuralBridge 

pub struct NeuralBridge { /* private fields */ }

Bridge for converting neural outputs to probabilistic constructs.

Handles numerical stability through epsilon clamping and normalization.

Implementations

impl NeuralBridge

pub fn new() -> Self

Create a new bridge with default epsilon (1e-8).

pub fn with_epsilon(epsilon: f64) -> Self

Create a bridge with custom epsilon.

pub fn to_ad_probabilities(&self, output: &NeuralOutput) -> Vec<ADProbability>

Convert softmax output to annotated disjunction probabilities.

Each probability is clamped to [epsilon, 1.0] for numerical stability.

pub fn batch_to_circuit_leaves( &self, outputs: &[NeuralOutput], ) -> Vec<Vec<CircuitLeaf>>

Convert batch of neural outputs to circuit leaf weights.

Returns a 2D structure: leaves[sample_idx][label_idx]

pub fn to_log_probabilities(&self, output: &NeuralOutput) -> Vec<f64>

Convert to log probabilities for numerical stability.

Log probabilities are used for:

• Computing NLL loss: -log(p_true)
• Avoiding underflow in product of many small probabilities

pub fn normalize(&self, output: &NeuralOutput) -> NeuralOutput

Normalize probabilities to sum to 1.0.

Useful when network outputs have small numerical errors.

pub fn extract_gradient_weights(&self, output: &NeuralOutput) -> Vec<f64>

Extract raw weights for gradient computation.

These weights are passed to the backward pass to compute gradients w.r.t. the neural network parameters.

pub fn probability_of(&self, output: &NeuralOutput, label: &str) -> Option<f64>

Compute the probability of a specific label.

pub fn log_probability_of( &self, output: &NeuralOutput, label: &str, ) -> Option<f64>

Compute the log probability of a specific label.

pub fn to_circuit_leaves_with_offset( &self, output: &NeuralOutput, variable_offset: usize, ) -> Vec<CircuitLeaf>

Create circuit leaves for a single sample with variable ID offset.

Used when multiple samples share a circuit structure but have different variable ID ranges.

Trait Implementations

impl Default for NeuralBridge

fn default() -> Self

Returns the “default value” for a type.