Introduce a common way to manage the coordination between ranking rules

This commit is contained in:
Loïc Lecrenier 2023-02-21 09:44:03 +01:00
parent 5065d8b0c1
commit ce0d1e0e13

View File

@ -0,0 +1,523 @@
use heed::RoTxn;
use roaring::RoaringBitmap;
use super::db_cache::DatabaseCache;
use super::resolve_query_graph::resolve_query_graph;
use super::QueryGraph;
use crate::new::graph_based_ranking_rule::GraphBasedRankingRule;
use crate::new::ranking_rule_graph::proximity::ProximityGraph;
use crate::new::words::Words;
// use crate::search::new::sort::Sort;
use crate::{Index, Result, TermsMatchingStrategy};
pub trait RankingRuleOutputIter<'transaction, Query> {
fn next_bucket(&mut self) -> Result<Option<RankingRuleOutput<Query>>>;
}
pub struct RankingRuleOutputIterWrapper<'transaction, Query> {
iter: Box<dyn Iterator<Item = Result<RankingRuleOutput<Query>>> + 'transaction>,
}
impl<'transaction, Query> RankingRuleOutputIterWrapper<'transaction, Query> {
pub fn new(
iter: Box<dyn Iterator<Item = Result<RankingRuleOutput<Query>>> + 'transaction>,
) -> Self {
Self { iter }
}
}
impl<'transaction, Query> RankingRuleOutputIter<'transaction, Query>
for RankingRuleOutputIterWrapper<'transaction, Query>
{
fn next_bucket(&mut self) -> Result<Option<RankingRuleOutput<Query>>> {
match self.iter.next() {
Some(x) => x.map(Some),
None => Ok(None),
}
}
}
pub trait RankingRuleQueryTrait: Sized + Clone + 'static {}
#[derive(Clone)]
pub struct PlaceholderQuery;
impl RankingRuleQueryTrait for PlaceholderQuery {}
impl RankingRuleQueryTrait for QueryGraph {}
pub trait RankingRule<'transaction, Query: RankingRuleQueryTrait> {
// TODO: add an update_candidates function to deal with distinct
// attributes?
fn start_iteration(
&mut self,
index: &Index,
txn: &'transaction RoTxn,
db_cache: &mut DatabaseCache<'transaction>,
universe: &RoaringBitmap,
query: &Query,
) -> Result<()>;
fn next_bucket(
&mut self,
index: &Index,
txn: &'transaction RoTxn,
db_cache: &mut DatabaseCache<'transaction>,
universe: &RoaringBitmap,
) -> Result<Option<RankingRuleOutput<Query>>>;
fn end_iteration(
&mut self,
index: &Index,
txn: &'transaction RoTxn,
db_cache: &mut DatabaseCache<'transaction>,
);
}
#[derive(Debug)]
pub struct RankingRuleOutput<Q> {
/// The query tree that must be used by the child ranking rule to fetch candidates.
pub query: Q,
/// The allowed candidates for the child ranking rule
pub candidates: RoaringBitmap,
}
#[allow(unused)]
pub fn get_start_universe<'transaction>(
index: &Index,
txn: &'transaction RoTxn,
db_cache: &mut DatabaseCache<'transaction>,
query_graph: &QueryGraph,
term_matching_strategy: TermsMatchingStrategy,
// filters: Filters,
// mut distinct: Option<D>,
) -> Result<RoaringBitmap> {
// NOTE:
//
// There is a performance problem when using `distinct` + exhaustive number of hits,
// especially for search that yield many results (many ~= almost all of the
// dataset).
//
// We'll solve it later. Maybe there are smart ways to go about it.
//
// For example, if there are millions of possible values for the distinct attribute,
// then we could just look at the documents which share any distinct attribute with
// another one, and remove the later docids them from the universe.
// => NO! because we don't know which one to remove, only after the sorting is done can we know it
// => this kind of computation can be done, but only in the evaluation of the number
// of hits for the documents that aren't returned by the search.
//
// `Distinct` otherwise should always be computed during
let universe = index.documents_ids(txn).unwrap();
// resolve the whole query tree to retrieve an exhaustive list of documents matching the query.
// NOTE: this is wrong
// Instead, we should only compute the documents corresponding to the last remaining
// word, 2-gram, and 3-gran.
// let candidates = resolve_query_graph(index, txn, db_cache, query_graph, &universe)?;
// Distinct should be lazy if placeholder?
//
// // because the initial_candidates should be an exhaustive count of the matching documents,
// // we precompute the distinct attributes.
// let initial_candidates = match &mut distinct {
// Some(distinct) => {
// let mut initial_candidates = RoaringBitmap::new();
// for c in distinct.distinct(candidates.clone(), RoaringBitmap::new()) {
// initial_candidates.insert(c?);
// }
// initial_candidates
// }
// None => candidates.clone(),
// };
Ok(/*candidates*/ universe)
}
pub fn execute_search<'transaction>(
index: &Index,
txn: &'transaction heed::RoTxn,
// TODO: ranking rules parameter
db_cache: &mut DatabaseCache<'transaction>,
universe: &RoaringBitmap,
query_graph: &QueryGraph,
// _from: usize,
// _length: usize,
) -> Result<Vec<u32>> {
let words = Words::new(TermsMatchingStrategy::Last);
// let sort = Sort::new(index, txn, "sort1".to_owned(), true)?;
let proximity = GraphBasedRankingRule::<ProximityGraph>::default();
// TODO: ranking rules given as argument
let mut ranking_rules: Vec<Box<dyn RankingRule<'transaction, QueryGraph>>> =
vec![Box::new(words), Box::new(proximity) /* Box::new(sort) */];
let ranking_rules_len = ranking_rules.len();
ranking_rules[0].start_iteration(index, txn, db_cache, universe, query_graph)?;
// TODO: parent_candidates could be used only during debugging?
let mut candidates = vec![RoaringBitmap::default(); ranking_rules_len];
candidates[0] = universe.clone();
let mut cur_ranking_rule_index = 0;
macro_rules! back {
() => {
candidates[cur_ranking_rule_index].clear();
ranking_rules[cur_ranking_rule_index].end_iteration(index, txn, db_cache);
if cur_ranking_rule_index == 0 {
break;
} else {
cur_ranking_rule_index -= 1;
}
};
}
let mut results = vec![];
// TODO: skip buckets when we want to start from an offset
while results.len() < 20 {
// The universe for this bucket is zero or one element, so we don't need to sort
// anything, just extend the results and go back to the parent ranking rule.
if candidates[cur_ranking_rule_index].len() <= 1 {
results.extend(&candidates[cur_ranking_rule_index]);
back!();
continue;
}
let Some(next_bucket) = ranking_rules[cur_ranking_rule_index].next_bucket(index, txn, db_cache, &candidates[cur_ranking_rule_index])? else {
back!();
continue;
};
candidates[cur_ranking_rule_index] -= &next_bucket.candidates;
if next_bucket.candidates.len() <= 1 {
// Only zero or one candidate, no need to sort through the child ranking rule.
results.extend(next_bucket.candidates);
continue;
} else {
// many candidates, give to next ranking rule, if any
if cur_ranking_rule_index == ranking_rules_len - 1 {
// TODO: don't extend too much, up to the limit only
results.extend(next_bucket.candidates);
} else {
cur_ranking_rule_index += 1;
candidates[cur_ranking_rule_index] = next_bucket.candidates.clone();
ranking_rules[cur_ranking_rule_index].start_iteration(
index,
txn,
db_cache,
&next_bucket.candidates,
&next_bucket.query,
)?;
}
}
}
Ok(results)
}
#[cfg(test)]
mod tests {
use std::fs::File;
use std::io::{BufRead, BufReader, Cursor, Seek};
use std::time::Instant;
use heed::EnvOpenOptions;
use super::{execute_search, get_start_universe};
use crate::documents::{DocumentsBatchBuilder, DocumentsBatchReader};
use crate::index::tests::TempIndex;
use crate::new::db_cache::DatabaseCache;
use crate::new::make_query_graph;
use crate::update::{IndexDocuments, IndexDocumentsConfig, IndexerConfig, Settings};
use crate::{Criterion, Index, Object, Search, TermsMatchingStrategy};
#[test]
fn execute_new_search() {
let index = TempIndex::new();
index
.add_documents(documents!([
{
"id": 7,
"text": "the super quick super brown fox jumps over",
},
{
"id": 8,
"text": "the super quick brown fox jumps over",
},
{
"id": 9,
"text": "the quick super brown fox jumps over",
},
{
"id": 10,
"text": "the quick brown fox jumps over",
},
{
"id": 11,
"text": "the quick brown fox jumps over the lazy dog",
},
{
"id": 12,
"text": "the quick brown cat jumps over the lazy dog",
},
]))
.unwrap();
let txn = index.read_txn().unwrap();
let mut db_cache = DatabaseCache::default();
let query_graph =
make_query_graph(&index, &txn, &mut db_cache, "the quick brown fox jumps over")
.unwrap();
println!("{}", query_graph.graphviz());
// TODO: filters + maybe distinct attributes?
let universe = get_start_universe(
&index,
&txn,
&mut db_cache,
&query_graph,
TermsMatchingStrategy::Last,
)
.unwrap();
println!("universe: {universe:?}");
let results =
execute_search(&index, &txn, &mut db_cache, &universe, &query_graph /* 0, 20 */)
.unwrap();
println!("{results:?}")
}
#[test]
fn search_movies_new() {
let mut options = EnvOpenOptions::new();
options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
let index = Index::new(options, "data_movies").unwrap();
let txn = index.read_txn().unwrap();
let primary_key = index.primary_key(&txn).unwrap().unwrap();
let primary_key = index.fields_ids_map(&txn).unwrap().id(primary_key).unwrap();
// loop {
// let start = Instant::now();
// let mut db_cache = DatabaseCache::default();
// let query_graph = make_query_graph(
// &index,
// &txn,
// &mut db_cache,
// "released from prison by the government",
// )
// .unwrap();
// // println!("{}", query_graph.graphviz());
// // TODO: filters + maybe distinct attributes?
// let universe = get_start_universe(
// &index,
// &txn,
// &mut db_cache,
// &query_graph,
// TermsMatchingStrategy::Last,
// )
// .unwrap();
// // println!("universe: {universe:?}");
// let results = execute_search(
// &index,
// &txn,
// &mut db_cache,
// &universe,
// &query_graph, /* 0, 20 */
// )
// .unwrap();
// let elapsed = start.elapsed();
// println!("{}us: {results:?}", elapsed.as_micros());
// }
let start = Instant::now();
let mut db_cache = DatabaseCache::default();
let query_graph =
make_query_graph(&index, &txn, &mut db_cache, "released from prison by the government")
.unwrap();
// println!("{}", query_graph.graphviz());
// TODO: filters + maybe distinct attributes?
let universe = get_start_universe(
&index,
&txn,
&mut db_cache,
&query_graph,
TermsMatchingStrategy::Last,
)
.unwrap();
// println!("universe: {universe:?}");
let results =
execute_search(&index, &txn, &mut db_cache, &universe, &query_graph /* 0, 20 */)
.unwrap();
let elapsed = start.elapsed();
let ids = index
.documents(&txn, results.iter().copied())
.unwrap()
.into_iter()
.map(|x| {
let obkv = &x.1;
let id = obkv.get(primary_key).unwrap();
let id: serde_json::Value = serde_json::from_slice(id).unwrap();
id.as_str().unwrap().to_owned()
})
.collect::<Vec<_>>();
println!("{}us: {results:?}", elapsed.as_micros());
println!("external ids: {ids:?}");
}
#[test]
fn search_movies_old() {
let mut options = EnvOpenOptions::new();
options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
let index = Index::new(options, "data_movies").unwrap();
let txn = index.read_txn().unwrap();
let start = Instant::now();
let mut s = Search::new(&txn, &index);
s.query("released from prison by the government");
s.terms_matching_strategy(TermsMatchingStrategy::Last);
// s.criterion_implementation_strategy(crate::CriterionImplementationStrategy::OnlySetBased);
let docs = s.execute().unwrap();
let elapsed = start.elapsed();
println!("{}us: {:?}", elapsed.as_micros(), docs.documents_ids);
}
#[test]
fn _settings_movies() {
let mut options = EnvOpenOptions::new();
options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
let index = Index::new(options, "data_movies").unwrap();
let mut wtxn = index.write_txn().unwrap();
// let primary_key = "id";
// let searchable_fields = vec!["title", "overview"];
// let filterable_fields = vec!["release_date", "genres"];
// let sortable_fields = vec[];
let config = IndexerConfig::default();
let mut builder = Settings::new(&mut wtxn, &index, &config);
builder.set_min_word_len_one_typo(5);
builder.set_min_word_len_two_typos(100);
// builder.set_primary_key(primary_key.to_owned());
// let searchable_fields = searchable_fields.iter().map(|s| s.to_string()).collect();
// builder.set_searchable_fields(searchable_fields);
// let filterable_fields = filterable_fields.iter().map(|s| s.to_string()).collect();
// builder.set_filterable_fields(filterable_fields);
builder.set_criteria(vec![Criterion::Words, Criterion::Proximity]);
// let sortable_fields = sortable_fields.iter().map(|s| s.to_string()).collect();
// builder.set_sortable_fields(sortable_fields);
builder.execute(|_| (), || false).unwrap();
}
// #[test]
fn _index_movies() {
let mut options = EnvOpenOptions::new();
options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
let index = Index::new(options, "data_movies").unwrap();
let mut wtxn = index.write_txn().unwrap();
let primary_key = "id";
let searchable_fields = vec!["title", "overview"];
let filterable_fields = vec!["release_date", "genres"];
// let sortable_fields = vec[];
let config = IndexerConfig::default();
let mut builder = Settings::new(&mut wtxn, &index, &config);
builder.set_primary_key(primary_key.to_owned());
let searchable_fields = searchable_fields.iter().map(|s| s.to_string()).collect();
builder.set_searchable_fields(searchable_fields);
let filterable_fields = filterable_fields.iter().map(|s| s.to_string()).collect();
builder.set_filterable_fields(filterable_fields);
builder.set_criteria(vec![Criterion::Words]);
// let sortable_fields = sortable_fields.iter().map(|s| s.to_string()).collect();
// builder.set_sortable_fields(sortable_fields);
builder.execute(|_| (), || false).unwrap();
let config = IndexerConfig::default();
let indexing_config = IndexDocumentsConfig::default();
let builder =
IndexDocuments::new(&mut wtxn, &index, &config, indexing_config, |_| (), || false)
.unwrap();
let documents = documents_from(
"/Users/meilisearch/Documents/milli2/benchmarks/datasets/movies.json",
"json",
);
let (builder, user_error) = builder.add_documents(documents).unwrap();
user_error.unwrap();
builder.execute().unwrap();
wtxn.commit().unwrap();
index.prepare_for_closing().wait();
}
fn documents_from(filename: &str, filetype: &str) -> DocumentsBatchReader<impl BufRead + Seek> {
let reader = File::open(filename)
.unwrap_or_else(|_| panic!("could not find the dataset in: {}", filename));
let reader = BufReader::new(reader);
let documents = match filetype {
"csv" => documents_from_csv(reader).unwrap(),
"json" => documents_from_json(reader).unwrap(),
"jsonl" => documents_from_jsonl(reader).unwrap(),
otherwise => panic!("invalid update format {:?}", otherwise),
};
DocumentsBatchReader::from_reader(Cursor::new(documents)).unwrap()
}
fn documents_from_jsonl(reader: impl BufRead) -> crate::Result<Vec<u8>> {
let mut documents = DocumentsBatchBuilder::new(Vec::new());
for result in serde_json::Deserializer::from_reader(reader).into_iter::<Object>() {
let object = result.unwrap();
documents.append_json_object(&object)?;
}
documents.into_inner().map_err(Into::into)
}
fn documents_from_json(reader: impl BufRead) -> crate::Result<Vec<u8>> {
let mut documents = DocumentsBatchBuilder::new(Vec::new());
documents.append_json_array(reader)?;
documents.into_inner().map_err(Into::into)
}
fn documents_from_csv(reader: impl BufRead) -> crate::Result<Vec<u8>> {
let csv = csv::Reader::from_reader(reader);
let mut documents = DocumentsBatchBuilder::new(Vec::new());
documents.append_csv(csv)?;
documents.into_inner().map_err(Into::into)
}
}