Struct EmbeddingMlpClassifier 

pub struct EmbeddingMlpClassifier {
    pub embedding: Vec<f32>,
    pub mlp: MlpClassifier,
    /* private fields */
}

A two-layer MLP classifier with a learnable token-embedding lookup.

This avoids materialising one-hot vectors for large vocabularies. The input is built from n_context_tokens token IDs whose embeddings are concatenated, followed by an optional continuous_dim vector of continuous features (e.g. geometric positions).

Fields

embedding: Vec<f32>

Token embedding table, row-major [vocab_size, embed_dim].

mlp: MlpClassifier

Underlying MLP head. Its input dimension is n_context_tokens * embed_dim + continuous_dim.

Implementations

impl EmbeddingMlpClassifier

pub fn new( vocab_size: usize, embed_dim: usize, n_context_tokens: usize, continuous_dim: usize, hidden_dim: usize, output_dim: usize, seed: u32, rope_theta: Option<f32>, ) -> Self

Create a fresh classifier with Xavier-like random weights.

pub fn input_dim(&self) -> usize

Total flat input dimension after embedding lookup and concatenation.

pub fn forward( &self, token_ids: &[u32], continuous: Option<&[f32]>, ) -> MlpForward

Forward pass for a single example.

pub fn predict(&self, token_ids: &[u32], continuous: Option<&[f32]>) -> usize

Predicted class for a single example.

pub fn loss( &self, token_ids: &[u32], continuous: Option<&[f32]>, target: usize, ) -> f32

Cross-entropy loss for a single example.

pub fn forward_batch( &self, token_ids: &[u32], continuous: Option<&[f32]>, batch_size: usize, ) -> MlpForwardBatch

Forward pass for a batch.

pub fn backward_batch( &self, token_ids: &[u32], continuous: Option<&[f32]>, targets: &[usize], batch_size: usize, ) -> (EmbeddingMlpGradients, f32)

Back-propagation for a batch.

Returns gradients averaged over the batch and the average cross-entropy loss. Embedding gradients are accumulated and averaged; the returned dembedding has the same shape as Self::embedding.

pub fn apply_sgd(&mut self, grad: &EmbeddingMlpGradients, lr: f32)

Apply a single SGD step using the supplied gradients.

Trait Implementations

impl Clone for EmbeddingMlpClassifier

fn clone(&self) -> EmbeddingMlpClassifier

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EmbeddingMlpClassifier

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

Formats the value using the given formatter.