Struct ZeroShotClassifier 

pub struct ZeroShotClassifier<B: Backend> { /* private fields */ }

Zero-shot classifier backed by a SensorLM model.

Implementations

impl<B: Backend> ZeroShotClassifier<B>

pub fn new<F>( model: SensorLMModel<B>, cfg: &ClassifierConfig, tokenize: F, ) -> Self

where F: Fn(&str) -> (Tensor<B, 2, Int>, Tensor<B, 2, Int>),

Construct the classifier and pre-compute class embeddings.

Arguments

• model – A trained SensorLM model.
• cfg – Classifier configuration.
• tokenize – A closure that converts a prompt string into (token_ids, attention_mask) tensors of shape (1, L).

pub fn predict(&self, sensor: Tensor<B, 3>) -> Vec<(usize, String, f32)>

Predict the class for each sensor sample in the batch.

Arguments

• sensor – (B, T, C) normalised sensor data.

Returns

A vector of B (class_index, class_name, similarity_score) tuples.

pub fn predict_topk( &self, sensor: Tensor<B, 3>, k: usize, ) -> Vec<Vec<(usize, String, f32)>>

Predict and return the top-k predictions per sample.