Struct RtModel 

pub struct RtModel { /* private fields */ }

A loaded RT model plus a content fingerprint (a stable hash of the model file bytes). The fingerprint keys the on-disk prediction cache (crate::rt_cache) so cached predictions are reused only for the exact model that produced them.

Implementations

impl RtModel

pub fn bundled() -> Result<Self>

The bundled CNN, with no runtime file or Python dependency.

pub fn cnn_info(&self) -> Option<&Value>

CNN provenance when available; legacy backends have no CNN metadata.

pub fn load(path: impl AsRef<Path>) -> Result<Self>

Load a distilled RT model, auto-detecting the backend by file magic: a leading UMLP => MLP, URTC => CNN, otherwise the gbdt-rs JSON format.

pub fn fingerprint(&self) -> u64

Stable content hash of the model file, used as the prediction-cache key.

pub fn predict_seq(&self, seq: &str) -> f64

Predict abstract RT for one sequence.

pub fn embed_many<S: AsRef<str> + Sync>( &self, seqs: &[S], ) -> Option<Vec<Vec<f64>>>

Penultimate-layer embeddings for fine-tuning the readout on-data. Only the MLP/CNN backends expose a usable embedding; the gbdt student returns None.

pub fn predict_embed_many<S: AsRef<str> + Sync>( &self, seqs: &[S], ) -> (Vec<f64>, Option<Vec<Vec<f64>>>)

Predict abstract RT AND return the penultimate embedding per sequence in a single pass, when the on-data RT fine-tune will consume both. Returns (preds, Some(embeds)) for backends with a usable embedding (MLP/CNN) and (preds, None) for the gbdt student. Byte-identical to calling predict_many and embed_many separately, but the CNN backend runs its forward pass once instead of twice over the same ~1.1M unique peptides.

pub fn predict_many<S: AsRef<str> + Sync>(&self, seqs: &[S]) -> Vec<f64>

Predict abstract RT for many sequences in one batch.