Possible acceleration of similarity.rs

[hrshdhgd]

Specifically

semsimian/src/similarity.rs

Lines 56 to 84 in 8d11f4c

	pub fn calculate_term_pairwise_information_content(
	closure_map: &HashMap<PredicateSetKey, HashMap<TermID, HashSet<TermID>>>,
	ic_map: &HashMap<PredicateSetKey, HashMap<TermID, f64>>,
	entity1: &HashSet<TermID>,
	entity2: &HashSet<TermID>,
	predicates: &Option<Vec<Predicate>>,
	) -> f64 {
	// At each iteration, it calculates the IC score using the calculate_max_information_content function,
	// and if the calculated IC score is greater than the current maximum IC (max_resnik_sim_e1_e2),
	// it updates the maximum IC value. Thus, at the end of the iterations,
	// max_resnik_sim_e1_e2 will contain the highest IC score among all the comparisons,
	// representing the best match between entity1 and entity2.
	let mut entity1_to_entity2_sum_resnik_sim = 0.0;
	for e1_term in entity1.iter() {
	let max_resnik_sim_e1_e2 = entity2
	.iter()
	.map(|e2_term| {
	calculate_max_information_content(closure_map, ic_map, e1_term, e2_term, predicates)
	})
	.fold(0.0, |max_ic: f64, (_max_ic_ancestors, ic)| max_ic.max(ic));
	entity1_to_entity2_sum_resnik_sim += max_resnik_sim_e1_e2;
	}
	// The final result will be the average Resnik similarity score between the two sets
	entity1_to_entity2_sum_resnik_sim / entity1.len() as f64
	}

Possible solution:

pub fn calculate_term_pairwise_information_content(
    closure_map: &HashMap<PredicateSetKey, HashMap<TermID, HashSet<TermID>>>,
    ic_map: &HashMap<PredicateSetKey, HashMap<TermID, f64>>,
    entity1: &HashSet<TermID>,
    entity2: &HashSet<TermID>,
    predicates: &Option<Vec<Predicate>>,
) -> f64 {
    let entity1_to_entity2_sum_resnik_sim: f64 = entity1.par_iter().map(|e1_term| {
        entity2.par_iter().map(|e2_term| {
            calculate_max_information_content(closure_map, ic_map, e1_term, e2_term, predicates)
        }).max_by(|a, b| a.partial_cmp(b).unwrap()).unwrap_or(0.0)
    }).sum();

    entity1_to_entity2_sum_resnik_sim / entity1.len() as f64
}