snowykami/meilisearch
1
0
Fork 0
mirror of https://github.com/meilisearch/meilisearch.git synced 2024-11-30 09:04:59 +08:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Compute the candidates but not by attribute

Browse Source
This commit is contained in:
Kerollmops 2020-07-13 18:16:05 +02:00
parent 6b14b20369
commit 576dd011a1
No known key found for this signature in database
GPG Key ID: 92ADA4E935E71FA4
1 changed files with 23 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
src/lib.rs
View File

@ -156,32 +156,31 @@ impl Index {
positions.push(union_positions.iter().collect()); positions.push(union_positions.iter().collect());
} }
let mut words_attributes_docids = Vec::new(); // We compute the docids candiate for these words (and derived words).
// We do a union between all the docids of each of the words and derived words,
// we got N unions (where N is the number of query words), we then intersect them.
// TODO we must store the words documents ids to avoid these unions.
let mut candidates = RoaringBitmap::new();
let number_of_attributes = self.number_of_attributes(rtxn)?.map_or(0, |n| n as u32); let number_of_attributes = self.number_of_attributes(rtxn)?.map_or(0, |n| n as u32);
for (i, derived_words) in words.iter().enumerate() {
for i in 0..number_of_attributes {
let mut intersect_docids: Option<RoaringBitmap> = None;
for derived_words in &words {
let mut union_docids = RoaringBitmap::new(); let mut union_docids = RoaringBitmap::new();
for (word, _) in derived_words { for (word, _) in derived_words {
// generate the key with the attribute number. for attr in 0..number_of_attributes {
let mut key = word.to_vec(); let mut key = word.to_vec();
key.extend_from_slice(&i.to_be_bytes()); key.extend_from_slice(&attr.to_be_bytes());
if let Some(right) = self.word_attribute_docids.get(rtxn, &key)? { if let Some(right) = self.word_attribute_docids.get(rtxn, &key)? {
union_docids.union_with(&right); union_docids.union_with(&right);
} }
} }
match &mut intersect_docids {
Some(left) => left.intersect_with(&union_docids),
None => intersect_docids = Some(union_docids),
} }
if i == 0 {
candidates = union_docids;
} else {
candidates.intersect_with(&union_docids);
} }
words_attributes_docids.push(intersect_docids);
} }
debug!("The documents you must find for each attribute: {:?}", words_attributes_docids); debug!("The candidates are {:?}", candidates);
debug!("Retrieving words positions took {:.02?}", before.elapsed()); debug!("Retrieving words positions took {:.02?}", before.elapsed());
// Returns the union of the same position for all the derived words. // Returns the union of the same position for all the derived words.
@ -202,7 +201,7 @@ impl Index {
let mut union_cache = HashMap::new(); let mut union_cache = HashMap::new();
let mut intersect_cache = HashMap::new(); let mut intersect_cache = HashMap::new();
// Returns `true` if there is documents in common between the two words and positions given. // Returns `true` if there is documents in common between the two words and positions given.
let mut contains_documents = |(lword, lpos), (rword, rpos), union_cache: &mut HashMap<_, _>, words_attributes_docids: &[_]| { let mut contains_documents = |(lword, lpos), (rword, rpos), union_cache: &mut HashMap<_, _>, candidates: &RoaringBitmap| {
let proximity = best_proximity::positions_proximity(lpos, rpos); let proximity = best_proximity::positions_proximity(lpos, rpos);
if proximity == 0 { return false } if proximity == 0 { return false }
@ -217,13 +216,9 @@ impl Index {
let lunion_docids = union_cache.get(&(lword, lpos)).unwrap(); let lunion_docids = union_cache.get(&(lword, lpos)).unwrap();
let runion_docids = union_cache.get(&(rword, rpos)).unwrap(); let runion_docids = union_cache.get(&(rword, rpos)).unwrap();
if proximity <= 7 { // We first check that the docids of these unions are part of the candidates.
let lattr = lpos / 1000; if lunion_docids.is_disjoint(candidates) { return false }
if let Some(docids) = &words_attributes_docids[lattr as usize] { if runion_docids.is_disjoint(candidates) { return false }
if lunion_docids.is_disjoint(&docids) { return false }
if runion_docids.is_disjoint(&docids) { return false }
}
}
!lunion_docids.is_disjoint(&runion_docids) !lunion_docids.is_disjoint(&runion_docids)
}) })
@ -231,7 +226,7 @@ impl Index {
let mut documents = Vec::new(); let mut documents = Vec::new();
let mut iter = BestProximity::new(positions); let mut iter = BestProximity::new(positions);
while let Some((proximity, mut positions)) = iter.next(|l, r| contains_documents(l, r, &mut union_cache, &words_attributes_docids)) { while let Some((proximity, mut positions)) = iter.next(|l, r| contains_documents(l, r, &mut union_cache, &candidates)) {
positions.sort_unstable(); positions.sort_unstable();
let same_prox_before = Instant::now(); let same_prox_before = Instant::now();
@ -289,12 +284,8 @@ impl Index {
} }
} }
// We achieve to find valid documents ids so we remove them from the candidate list. // We achieve to find valid documents ids so we remove them from the candidates list.
for docids in &mut words_attributes_docids { candidates.difference_with(&same_proximity_union);
if let Some(docids) = docids {
docids.difference_with(&same_proximity_union);
}
}
documents.push(same_proximity_union); documents.push(same_proximity_union);