snowykami/meilisearch
1
0
Fork 0
mirror of https://github.com/meilisearch/meilisearch.git synced 2024-11-23 10:37:41 +08:00
Code Issues Actions 8 Packages Projects Releases Wiki Activity

Reduce the amount of branches when query tree flattened

Browse Source
This commit is contained in:
Clément Renault 2021-03-29 16:25:14 +02:00 committed by many
parent 7ff4a2a708
commit 361193099f
No known key found for this signature in database
GPG Key ID: 2CEF23B75189EACA
1 changed files with 49 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

83
milli/src/search/criteria/attribute.rs
View File

@ -1,3 +1,4 @@
use std::cmp;
use std::collections::{BTreeMap, HashMap, btree_map}; use std::collections::{BTreeMap, HashMap, btree_map};
use std::mem::take; use std::mem::take;
@ -15,7 +16,7 @@ pub struct Attribute<'t> {
candidates: Option<RoaringBitmap>, candidates: Option<RoaringBitmap>,
bucket_candidates: RoaringBitmap, bucket_candidates: RoaringBitmap,
parent: Box<dyn Criterion + 't>, parent: Box<dyn Criterion + 't>,
flattened_query_tree: Option<Vec<Vec<Query>>>, flattened_query_tree: Option<Vec<Vec<Vec<Query>>>>,
current_buckets: Option<btree_map::IntoIter<u64, RoaringBitmap>>, current_buckets: Option<btree_map::IntoIter<u64, RoaringBitmap>>,
} }
@ -115,33 +116,43 @@ impl<'t> Criterion for Attribute<'t> {
fn linear_compute_candidates( fn linear_compute_candidates(
ctx: &dyn Context, ctx: &dyn Context,
branches: &Vec<Vec<Query>>, branches: &Vec<Vec<Vec<Query>>>,
allowed_candidates: &RoaringBitmap, allowed_candidates: &RoaringBitmap,
) -> anyhow::Result<BTreeMap<u64, RoaringBitmap>> ) -> anyhow::Result<BTreeMap<u64, RoaringBitmap>>
{ {
fn compute_candidate_rank(branches: &Vec<Vec<Query>>, words_positions: HashMap<String, RoaringBitmap>) -> u64 { fn compute_candidate_rank(branches: &Vec<Vec<Vec<Query>>>, words_positions: HashMap<String, RoaringBitmap>) -> u64 {
let mut min_rank = u64::max_value(); let mut min_rank = u64::max_value();
for branch in branches { for branch in branches {
let branch_len = branch.len(); let branch_len = branch.len();
let mut branch_rank = Vec::with_capacity(branch_len); let mut branch_rank = Vec::with_capacity(branch_len);
for Query { prefix, kind } in branch { for derivates in branch {
// find the best position of the current word in the document. let mut position = None;
let position = match kind { for Query { prefix, kind } in derivates {
QueryKind::Exact { word, .. } => { // find the best position of the current word in the document.
if *prefix { let current_position = match kind {
word_derivations(word, true, 0, &words_positions) QueryKind::Exact { word, .. } => {
.flat_map(|positions| positions.iter().next()).min() if *prefix {
} else { word_derivations(word, true, 0, &words_positions)
words_positions.get(word) .flat_map(|positions| positions.iter().next()).min()
.map(|positions| positions.iter().next()) } else {
.flatten() words_positions.get(word)
} .map(|positions| positions.iter().next())
}, .flatten()
QueryKind::Tolerant { typo, word } => { }
word_derivations(word, *prefix, *typo, &words_positions) },
.flat_map(|positions| positions.iter().next()).min() QueryKind::Tolerant { typo, word } => {
}, word_derivations(word, *prefix, *typo, &words_positions)
}; .flat_map(|positions| positions.iter().next()).min()
},
};
match (position, current_position) {
(Some(p), Some(cp)) => position = Some(cmp::min(p, cp)),
(None, Some(cp)) => position = Some(cp),
_ => (),
}
}
// if a position is found, we add it to the branch score, // if a position is found, we add it to the branch score,
// otherwise the branch is considered as unfindable in this document and we break. // otherwise the branch is considered as unfindable in this document and we break.
@ -194,10 +205,10 @@ fn linear_compute_candidates(
} }
// TODO can we keep refs of Query // TODO can we keep refs of Query
fn flatten_query_tree(query_tree: &Operation) -> Vec<Vec<Query>> { fn flatten_query_tree(query_tree: &Operation) -> Vec<Vec<Vec<Query>>> {
use crate::search::criteria::Operation::{And, Or, Consecutive}; use crate::search::criteria::Operation::{And, Or, Consecutive};
fn and_recurse(head: &Operation, tail: &[Operation]) -> Vec<Vec<Query>> { fn and_recurse(head: &Operation, tail: &[Operation]) -> Vec<Vec<Vec<Query>>> {
match tail.split_first() { match tail.split_first() {
Some((thead, tail)) => { Some((thead, tail)) => {
let tail = and_recurse(thead, tail); let tail = and_recurse(thead, tail);
@ -215,13 +226,17 @@ fn flatten_query_tree(query_tree: &Operation) -> Vec<Vec<Query>> {
} }
} }
fn recurse(op: &Operation) -> Vec<Vec<Query>> { fn recurse(op: &Operation) -> Vec<Vec<Vec<Query>>> {
match op { match op {
And(ops) | Consecutive(ops) => { And(ops) | Consecutive(ops) => {
ops.split_first().map_or_else(Vec::new, |(h, t)| and_recurse(h, t)) ops.split_first().map_or_else(Vec::new, |(h, t)| and_recurse(h, t))
}, },
Or(_, ops) => ops.into_iter().map(recurse).flatten().collect(), Or(_, ops) => if ops.iter().all(|op| op.query().is_some()) {
Operation::Query(query) => vec![vec![query.clone()]], vec![vec![ops.iter().flat_map(|op| op.query()).cloned().collect()]]
} else {
ops.into_iter().map(recurse).flatten().collect()
},
Operation::Query(query) => vec![vec![vec![query.clone()]]],
} }
} }
@ -256,19 +271,19 @@ mod tests {
]); ]);
let expected = vec![ let expected = vec![
vec![Query { prefix: false, kind: QueryKind::exact(S("manythefish")) }], vec![vec![Query { prefix: false, kind: QueryKind::exact(S("manythefish")) }]],
vec![ vec![
Query { prefix: false, kind: QueryKind::exact(S("manythe")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("manythe")) }],
Query { prefix: false, kind: QueryKind::exact(S("fish")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("fish")) }],
], ],
vec![ vec![
Query { prefix: false, kind: QueryKind::exact(S("many")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("many")) }],
Query { prefix: false, kind: QueryKind::exact(S("thefish")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("thefish")) }],
], ],
vec![ vec![
Query { prefix: false, kind: QueryKind::exact(S("many")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("many")) }],
Query { prefix: false, kind: QueryKind::exact(S("the")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("the")) }],
Query { prefix: false, kind: QueryKind::exact(S("fish")) }, vec![Query { prefix: false, kind: QueryKind::exact(S("fish")) }],
], ],
]; ];