Struct CorefEvaluation 

pub struct CorefEvaluation {
    pub muc: CorefScores,
    pub b_cubed: CorefScores,
    pub ceaf_e: CorefScores,
    pub ceaf_m: CorefScores,
    pub lea: CorefScores,
    pub blanc: CorefScores,
    pub conll_f1: f64,
    pub chain_stats: Option<CorefChainStats>,
    pub zero_anaphor: Option<ZeroAnaphorEvaluation>,
}

Complete coreference evaluation results.

Fields

muc: CorefScores

MUC metric.

b_cubed: CorefScores

B³ metric.

ceaf_e: CorefScores

CEAF entity-based (phi4).

ceaf_m: CorefScores

CEAF mention-based (phi3).

lea: CorefScores

LEA metric.

blanc: CorefScores

BLANC metric.

conll_f1: f64

CoNLL F1 (average of MUC, B³, CEAFe).

chain_stats: Option<CorefChainStats>

Chain-length stratified diagnostics.

zero_anaphor: Option<ZeroAnaphorEvaluation>

Zero-anaphor (empty-node) evaluation, if zeros are present.

Implementations

impl CorefEvaluation

pub fn compute(predicted: &[CorefChain], gold: &[CorefChain]) -> CorefEvaluation

Compute the full metric bundle.

use anno::core::coref::{CorefChain, Mention};
use anno::metrics::coref_metrics::CorefEvaluation;

let gold = vec![CorefChain::new(vec![
    Mention::new("John", 0, 4),
    Mention::new("he", 10, 12),
])];
let eval = CorefEvaluation::compute(&gold, &gold);
assert!((eval.conll_f1 - 1.0).abs() < 1e-9);
assert!((eval.muc.f1 - 1.0).abs() < 1e-9);

pub fn compute_without_singletons( predicted: &[CorefChain], gold: &[CorefChain], ) -> CorefEvaluation

Compute the full metric bundle with singletons excluded.

CRAC 2022-2025 shared tasks use this as the primary scoring mode: chains with only one mention are removed from both predicted and gold before computing B3, CEAF, LEA, and CoNLL scores.

Scores typically differ from the singleton-included variant because singletons inflate B3 precision/recall when correctly matched but contribute no signal to MUC (which already ignores them).

use anno::core::coref::{CorefChain, Mention};
use anno::metrics::coref_metrics::CorefEvaluation;

let gold = vec![
    CorefChain::new(vec![
        Mention::new("John", 0, 4),
        Mention::new("he", 10, 12),
    ]),
    CorefChain::singleton(Mention::new("Paris", 20, 25)),
];
let with = CorefEvaluation::compute(&gold, &gold);
let without = CorefEvaluation::compute_without_singletons(&gold, &gold);
// B3 F1 is 1.0 either way for identical input, but the chain_stats differ.
assert!(with.chain_stats.as_ref().unwrap().singleton_count == 1);
assert!(without.chain_stats.as_ref().unwrap().singleton_count == 0);

pub fn all_f1_scores(&self) -> Vec<f64>

Extract per-metric F1 scores.

pub fn average_f1(&self) -> f64

Average F1 across all included metrics (diagnostic).

pub fn f1_std_dev(&self) -> f64

Standard deviation of F1 scores across metrics.

Trait Implementations

impl Clone for CorefEvaluation

fn clone(&self) -> CorefEvaluation

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CorefEvaluation

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

Formats the value using the given formatter.

impl Default for CorefEvaluation

fn default() -> CorefEvaluation

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for CorefEvaluation

fn deserialize<__D>( __deserializer: __D, ) -> Result<CorefEvaluation, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for CorefEvaluation

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

Formats the value using the given formatter.

impl Serialize for CorefEvaluation

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

where __S: Serializer,

Serialize this value into the given Serde serializer.