Struct EvalSystem 

pub struct EvalSystem { /* private fields */ }

Unified evaluation system - single entry point for all evaluations.

Implementations

impl EvalSystem

pub fn new() -> Self

Create a new unified evaluation system.

pub fn with_tasks(self, tasks: Vec<Task>) -> Self

Set tasks to evaluate.

pub fn with_datasets(self, datasets: Vec<DatasetId>) -> Self

Set datasets to use.

pub fn with_backends(self, backends: Vec<String>) -> Self

Set backends to test.

pub fn with_backend_names(self, backends: Vec<BackendName>) -> Self

Set backends using type-safe BackendName enum.

pub fn with_max_examples(self, max: Option<usize>) -> Self

Set maximum examples per dataset.

Pass None to remove limit (evaluate all examples).

pub fn add_task(self, task: Task) -> Self

Add a task to evaluate.

pub fn add_dataset(self, dataset: DatasetId) -> Self

Add a dataset to use.

pub fn add_backend(self, backend: String) -> Self

Add a backend to test.

pub fn add_backend_name(self, backend: BackendName) -> Self

Add a backend using type-safe BackendName enum.

pub fn with_seed(self, seed: u64) -> Self

Set random seed.

pub fn with_bias_analysis(self, enable: bool) -> Self

Enable bias analysis.

pub fn with_bias_config(self, config: BiasDatasetConfig) -> Self

Set bias evaluation configuration.

pub fn with_calibration(self, enable: bool) -> Self

Enable calibration analysis.

pub fn with_data_quality(self, enable: bool) -> Self

Enable data quality checks.

pub fn with_model(self, model: Box<dyn Model>, name: Option<String>) -> Self

Set model to evaluate (for bias/calibration that need model instance).

pub fn with_coref_resolver(self, resolver: Box<dyn CoreferenceResolver>) -> Self

Set coreference resolver to evaluate.

This allows evaluating coreference resolvers (e.g., TrainedBoxCorefResolver from matryoshka-box) using anno’s evaluation infrastructure.

Example

use matryoshka_box_inference::trained_resolver::TrainedBoxCorefResolver;
use anno_eval::eval::unified_evaluator::EvalSystem;
use anno_eval::eval::task_mapping::Task;

let resolver = TrainedBoxCorefResolver::new(trained_boxes, config);
let results = EvalSystem::new()
    .with_coref_resolver(Box::new(resolver))
    .with_tasks(vec![Task::Coreference { metrics: vec![] }])
    .run()?;

pub fn run(self) -> Result<UnifiedEvalResults>

Run all enabled evaluations.

Trait Implementations

impl Default for EvalSystem

fn default() -> Self

Returns the “default value” for a type.