Struct ZeroShotSelector 

pub struct ZeroShotSelector { /* private fields */ }

Zero-shot optimizer selector: an offline-fitted meta-model plus the meta-feature extractor it consumes.

The selector is fit once on a meta-dataset of MetaExamples (see ZeroShotSelector::fit) and thereafter answers ZeroShotSelector::recommend queries for brand-new tasks with no further training.

Implementations

impl ZeroShotSelector

pub fn new(config: ZeroShotConfig) -> Result<Self>

Create an unfitted selector with the given configuration.

pub fn with_default_config() -> Result<Self>

Create an unfitted selector with the default configuration.

pub fn is_fitted(&self) -> bool

Whether ZeroShotSelector::fit has been called successfully.

pub fn config(&self) -> &ZeroShotConfig

The active configuration.

pub fn extract_meta_features<P>(&self, probe: &P) -> Result<MetaFeatures>

where P: TaskProbe + ?Sized,

Extract the meta-feature vector for a task by lightweight probing.

This performs only gradient evaluations (no optimization steps on the task) and is deterministic: the random curvature directions are drawn from a generator seeded by ZeroShotConfig::seed, and the noise estimate uses index-seeded gradient samples, so repeated calls return identical features.

pub fn fit(&mut self, examples: &[MetaExample]) -> Result<()>

Fit the meta-model on an offline meta-dataset.

This learns the standardization statistics, the multinomial logistic-regression classifier (softmax cross-entropy, full-batch gradient descent with L2), and the linear log10(learning_rate) regressor (MSE, full-batch gradient descent with L2).

pub fn recommend<P>(&self, probe: &P) -> Result<OptimizerRecommendation>

where P: TaskProbe + ?Sized,

Recommend an optimizer and hyper-parameters for a brand-new task.

Extracts meta-features from the probe, standardizes them with the fitted statistics, and runs the classifier + regressor. Returns an error if the selector has not been fit yet.

pub fn predict( &self, features: &MetaFeatures, ) -> Result<OptimizerRecommendation>

Predict directly from an already-extracted meta-feature vector.

Returns an error if the selector has not been fit yet.

pub fn save_json(&self, path: &Path) -> Result<()>

Serialize the fitted selector (including its learned parameters) to JSON on disk.

pub fn load_json(path: &Path) -> Result<Self>

Load a previously-saved selector from JSON on disk.

The deserialized value is validated before it is handed back: JSON is structurally permissive, so a file that parses can still carry a weight matrix of the wrong shape, a zero/negative feature standard deviation (which would divide by zero during standardization) or a non-finite parameter. Those used to sail through and panic — or silently produce garbage predictions — at the first predict call.

Errors

Returns Err on I/O failure, on malformed JSON, or when Self::validate_parameters rejects the loaded parameters.

pub fn validate_parameters(&self) -> Result<()>

Check that the learned parameters are internally consistent and usable.

Verifies the configuration, every array shape against NUM_FEATURES / NUM_CLASSES, that no parameter is NaN/infinite, and that every feature standard deviation is strictly positive.

Errors

Returns Err describing the first inconsistency found.

Trait Implementations

impl Clone for ZeroShotSelector

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ZeroShotSelector

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ZeroShotSelector

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ZeroShotSelector

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.