Struct AttributeReconstructionObjective 

pub struct AttributeReconstructionObjective { /* private fields */ }

A simple multi-layer perceptron for attribute reconstruction.

Architecture: input_dim → hidden_dim → … → input_dim (tanh between hidden layers, linear output). No back-propagation is implemented here; this struct provides the forward pass for inference / loss computation.

Implementations

impl AttributeReconstructionObjective

pub fn new(input_dim: usize, config: AttrReconConfig, seed: u64) -> Result<Self>

Build a new MLP for input_dim-dimensional features.

Layers are randomly initialised. Use the same seed for reproducibility.

Errors

Returns GraphError::InvalidParameter if input_dim == 0 or n_layers == 0.

pub fn forward(&self, features: &[Vec<f64>]) -> Result<Vec<Vec<f64>>>

Run a forward pass for each node’s feature vector.

Arguments

• features – n_nodes × input_dim feature matrix.

Returns

Reconstructed feature matrix of the same shape.

Errors

Returns GraphError::InvalidParameter if any feature vector has length ≠ input_dim.

pub fn mse_loss( &self, predicted: &[Vec<f64>], target: &[Vec<f64>], ) -> Result<f64>

Mean squared error between predicted and target.

Errors

Returns GraphError::InvalidParameter if shapes differ.

pub fn config(&self) -> &AttrReconConfig

Access the underlying config.