caio.co/de/cantine


Add support for customizing the scores 💬 by Caio 4 years ago (log)
Kind of a `tique::top_collector::custom_score` copy pasta with cleaned
up trait requirements and added support for type-based ordering.

Blob tique/src/conditional_collector/top_collector.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
use std::marker::PhantomData;

use tantivy::{
    collector::{Collector, SegmentCollector},
    DocAddress, DocId, Result, Score, SegmentLocalId, SegmentReader,
};

use super::{
    topk::{TopK, TopKProvider},
    traits::{CheckCondition, ConditionForSegment},
};

pub struct TopCollector<T, P, CF> {
    limit: usize,
    condition_for_segment: CF,
    _score: PhantomData<T>,
    _provider: PhantomData<P>,
}

impl<T, P, CF> TopCollector<T, P, CF>
where
    T: PartialOrd,
    P: TopKProvider<T>,
    CF: ConditionForSegment<T>,
{
    pub fn new(limit: usize, condition_for_segment: CF) -> Self {
        if limit < 1 {
            panic!("Limit must be greater than 0");
        }
        TopCollector {
            limit,
            condition_for_segment,
            _score: PhantomData,
            _provider: PhantomData,
        }
    }
}

impl<P, CF> Collector for TopCollector<Score, P, CF>
where
    P: 'static + Send + Sync + TopKProvider<Score>,
    CF: Sync + ConditionForSegment<Score>,
{
    type Fruit = CollectionResult<Score>;
    type Child = TopSegmentCollector<Score, P::Child, CF::Type>;

    fn requires_scoring(&self) -> bool {
        true
    }

    fn merge_fruits(&self, children: Vec<Self::Fruit>) -> Result<Self::Fruit> {
        Ok(P::merge_many(self.limit, children))
    }

    fn for_segment(
        &self,
        segment_id: SegmentLocalId,
        reader: &SegmentReader,
    ) -> Result<Self::Child> {
        Ok(TopSegmentCollector::new(
            segment_id,
            P::new_topk(self.limit),
            self.condition_for_segment.for_segment(reader),
        ))
    }
}

pub struct TopSegmentCollector<T, K, C> {
    total: usize,
    visited: usize,
    segment_id: SegmentLocalId,
    topk: K,
    condition: C,
    _marker: PhantomData<T>,
}

impl<T, K, C> TopSegmentCollector<T, K, C>
where
    T: Copy,
    K: TopK<T, DocId>,
    C: CheckCondition<T>,
{
    pub fn new(segment_id: SegmentLocalId, topk: K, condition: C) -> Self {
        Self {
            total: 0,
            visited: 0,
            segment_id,
            topk,
            condition,
            _marker: PhantomData,
        }
    }

    #[cfg(test)]
    fn into_topk(self) -> K {
        self.topk
    }

    pub fn collect(&mut self, doc: DocId, score: T) {
        self.total += 1;
        if self
            .condition
            .check(self.segment_id, doc, score, K::ASCENDING)
        {
            self.visited += 1;
            self.topk.visit(score, doc);
        }
    }

    pub fn into_collection_result(self) -> CollectionResult<T> {
        let segment_id = self.segment_id;
        let items = self
            .topk
            .into_vec()
            .into_iter()
            .map(|(score, doc)| (score, DocAddress(segment_id, doc)))
            .collect();

        // XXX This is unsorted. It's ok because we sort during
        // merge, but using the same time to mean two things is
        // rather confusing
        CollectionResult {
            total: self.total,
            visited: self.visited,
            items,
        }
    }
}

impl<K, C> SegmentCollector for TopSegmentCollector<Score, K, C>
where
    K: 'static + TopK<Score, DocId>,
    C: CheckCondition<Score>,
{
    type Fruit = CollectionResult<Score>;

    fn collect(&mut self, doc: DocId, score: Score) {
        TopSegmentCollector::collect(self, doc, score)
    }

    fn harvest(self) -> Self::Fruit {
        TopSegmentCollector::into_collection_result(self)
    }
}

#[derive(Debug)]
pub struct CollectionResult<T> {
    pub total: usize,
    pub visited: usize,
    pub items: Vec<(T, DocAddress)>,
}

#[cfg(test)]
mod tests {

    use super::*;
    use crate::conditional_collector::{
        topk::{AscendingTopK, DescendingTopK},
        Ascending, Descending,
    };

    use tantivy::{query::TermQuery, schema, Document, Index, Result, Term};

    #[test]
    fn condition_is_checked() {
        const LIMIT: usize = 4;

        let mut nil_collector = TopSegmentCollector::new(0, AscendingTopK::new(LIMIT), false);

        let mut top_collector = TopSegmentCollector::new(0, AscendingTopK::new(LIMIT), true);

        let condition = |_sid, doc, _score, _asc| doc % 2 == 1;

        let mut just_odds = TopSegmentCollector::new(0, AscendingTopK::new(LIMIT), condition);

        for i in 0..4 {
            nil_collector.collect(i, 420.0);
            top_collector.collect(i, 420.0);
            just_odds.collect(i, 420.0);
        }

        assert_eq!(0, nil_collector.harvest().items.len());
        assert_eq!(4, top_collector.harvest().items.len());

        // Verify that the collected items respect the condition
        let result = just_odds.harvest();
        assert_eq!(4, result.total);
        assert_eq!(2, result.items.len());
        for (score, doc) in result.items {
            let DocAddress(seg_id, doc_id) = doc;
            assert!(condition(seg_id, doc_id, score, true))
        }
    }

    fn check_segment_collector<K, C>(
        topk: K,
        condition: C,
        input: Vec<(Score, DocId)>,
        wanted: Vec<(Score, DocId)>,
    ) where
        K: TopK<Score, DocId> + 'static,
        C: CheckCondition<Score>,
    {
        let mut collector = TopSegmentCollector::new(0, topk, condition);

        for (score, id) in input {
            collector.collect(id, score);
        }

        assert_eq!(wanted, collector.into_topk().into_sorted_vec());
    }

    #[test]
    fn collection_with_a_marker_smoke() {
        // XXX property test maybe? Essentially we are creating
        // a Vec<(Score, DocId)> sorted as `Scored` would,
        // then we pick an arbitrary position to pivot and
        // expect the DescendingTopK to pick everything below
        // and the AscendingTopK to pick everything above
        let marker = (0.5, DocAddress(0, 4));

        check_segment_collector(
            DescendingTopK::new(10),
            marker,
            vec![
                // Every doc with a higher score has appeared already
                (0.6, 7),
                (0.7, 5),
                // Docs with the same score, but lower id too
                (0.5, 3),
                (0.5, 2),
                // [pivot] And, of course, the same doc should not be collected
                (0.5, 4),
                // Lower scores are in
                (0.0, 1),
                // Same score but higher doc, too
                (0.5, 6),
            ],
            vec![(0.5, 6), (0.0, 1)],
        );

        check_segment_collector(
            AscendingTopK::new(10),
            marker,
            vec![
                // Every doc with a higher score should be picked
                (0.6, 7),
                (0.7, 5),
                // Same score but lower id as well
                (0.5, 3),
                (0.5, 2),
                // [pivot] The same doc should not be collected
                (0.5, 4),
                // Docs with lower scores are discarded
                (0.0, 1),
                // Same score but higher doc is discaraded too
                (0.5, 6),
            ],
            vec![(0.5, 2), (0.5, 3), (0.6, 7), (0.7, 5)],
        );
    }

    #[test]
    fn collection_ordering_integration() -> Result<()> {
        let mut builder = schema::SchemaBuilder::new();

        let text_field = builder.add_text_field("text", schema::TEXT);

        let index = Index::create_in_ram(builder.build());
        let mut writer = index.writer_with_num_threads(1, 3_000_000)?;

        let add_doc = |text: &str| {
            let mut doc = Document::new();
            doc.add_text(text_field, text);
            writer.add_document(doc);
        };

        const NUM_DOCS: usize = 3;
        add_doc("the first doc is simple");
        add_doc("the second doc is a bit larger");
        add_doc("and the third document is rubbish");

        writer.commit()?;

        let reader = index.reader()?;
        let searcher = reader.searcher();

        let collector_asc = TopCollector::<_, Ascending, _>::new(NUM_DOCS, true);
        let collector_desc = TopCollector::<_, Descending, _>::new(NUM_DOCS, true);

        // Query for "the", which matches all docs and yields
        // a distinct score for each
        let query = TermQuery::new(
            Term::from_field_text(text_field, "the"),
            schema::IndexRecordOption::WithFreqsAndPositions,
        );
        let (asc, desc) = searcher.search(&query, &(collector_asc, collector_desc))?;

        assert_eq!(NUM_DOCS, asc.items.len());
        assert_eq!(NUM_DOCS, desc.items.len());

        let asc_scores = asc
            .items
            .iter()
            .map(|(score, _doc)| score)
            .collect::<Vec<_>>();

        let mut prev = None;
        for score in &asc_scores {
            if let Some(previous) = prev {
                assert!(previous < score, "The scores should be ascending");
            }
            prev = Some(score)
        }

        let mut desc_scores = desc
            .items
            .iter()
            .map(|(score, _doc)| score)
            .collect::<Vec<_>>();

        desc_scores.reverse();
        assert_eq!(asc_scores, desc_scores);

        Ok(())
    }
}